Inference device, inference method, inference program, model generation method, inference service providing system, inference service providing method, and inference service providing program

Abstract

The present invention assists the work of designing or selecting the chemical structure of a lipid molecule making up a particle including an active substance. This inference device comprises an acquiring unit for acquiring input data including the chemical structure information of at least a lipid molecule, and a trained model generated by performing a learning process with respect to a learning model that associates input data including the chemical structure information of at least a lipid molecule with the cell transfection efficiency and / or cell survival rate of the active substance included in the particle containing the lipid molecule. The trained model infers the transfection efficiency and / or cell survival rate associated with input data newly acquired by the acquiring unit.

Description

[Technical Field]

[0001] The present disclosure relates to an inference device, an inference method, an inference program, a model generation method, an inference service providing system, an inference service providing method, and an inference service providing program.

[0002] Drug delivery systems (DDSs) using particles containing lipid molecules are known for the highly efficient introduction of active substances such as nucleic acids into cells. In these systems, the active substance is encapsulated in particles containing lipid molecules to form complex particles, and the active substance is then introduced (transfected) into cells via these complex particles. Such DDSs are used not only for transfection of cells in vivo by administration to the body, but also for transfection of cells outside the body (in vitro, in situ, or ex vivo). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2016 / 021683 [Patent Document 2] International Publication No. 2019 / 131839 [Patent Document 3] International Publication No. 2020 / 032184 Summary of the Invention [Problem to be solved by the invention]

[0004] On the other hand, the design and selection of the chemical structures of lipid molecules that constitute particles containing active substances are generally performed manually, so the design and selection of appropriate lipid molecules for a specific purpose largely depends on the experience and know-how of the skilled person. Furthermore, since lipid molecules with the designed or selected chemical structures are experimentally evaluated and the design and selection are repeatedly redone based on the evaluation results, it takes time to search for more appropriate lipid molecular structures. Furthermore, simply searching for lipid molecules suitable for a limited number of active substances or purposes makes it difficult to accumulate know-how for designing or selecting lipid molecular structures suitable for a variety of active substances and purposes.

[0005] The present disclosure is intended to aid in the design or selection of chemical structures of lipid molecules that make up particles that encase active substances. [Means for solving the problem]

[0006] In view of the above problems, the present inventors have conducted extensive research and have found that it is possible to generate a learning model based on data such as chemical structure information of lipid molecules, transfection efficiency and / or cell survival rate, and to use this model to infer chemical structure information of lipid molecules, transfection efficiency and / or cell survival rate, etc.

[0007] That is, the present disclosure provides the following: [1] An acquisition unit that acquires input data including at least chemical structure information of lipid molecules; a trained model generated by performing a training process on a training model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of an active substance encapsulated in a particle containing the lipid molecules and / or the cell survival rate; An inference device, wherein the trained model infers transfection efficiency and / or cell viability associated with input data newly acquired by the acquisition unit. [2] The inference device described in [1], wherein the transfection efficiency and / or cell viability used when the learning process is performed is calculated from the measurement results obtained by introducing into cells an active substance encapsulated in particles containing lipid molecules having the chemical structure information used when the learning process is performed. [3] The inference device described in [2], wherein the trained model is generated by updating the model parameters of the training model so that the output when input data including chemical structure information of at least the lipid molecules is input to the training model approaches the transfection efficiency and / or cell survival rate calculated from the measurement results. "4" The inference device described in [1], wherein the acquisition unit performs predetermined preprocessing on the newly acquired input data, and the trained model infers transfection efficiency and / or cell survival rate associated with the preprocessed input data. [5] An acquisition step of acquiring input data including at least chemical structure information of lipid molecules; an execution step of executing a trained model generated by performing a training process on a training model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of an active substance encapsulated in a particle containing the lipid molecules and / or the cell survival rate, An inference method in which the execution step executes the trained model to infer transfection efficiency and / or cell viability associated with the newly acquired input data in the acquisition step. [6] An acquisition step of acquiring input data including at least chemical structure information of lipid molecules; an execution step of executing a trained model generated by performing a training process on a training model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of an active substance encapsulated in a particle containing the lipid molecules and / or the cell survival rate, The execution step is an inference program that executes the trained model to infer transfection efficiency and / or cell viability associated with the newly acquired input data in the acquisition step. [7] A model generation method for generating a trained model by performing a learning process on a learning model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency and / or cell survival rate of an active substance encapsulated in a particle containing the lipid molecules. [8] An acquisition unit that acquires input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; A trained model is generated by performing a training process on a training model that associates input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules, The trained model infers chemical structure information of lipid molecules associated with input data newly acquired by the acquisition unit. [9] The inference device described in [8], wherein the input data used when the learning process is performed includes the transfection efficiency of the active substance encapsulated in particles containing designed or selected lipid molecules into cells and / or the cell survival rate, calculated from measurement results obtained by introducing the active substance encapsulated in particles containing the lipid molecules into cells.

[10] The inference device described in [8], wherein the trained model is generated by updating the model parameters of the training model so that the output when input data including the prerequisites is input to the training model approaches the chemical structure information of the lipid molecules used when the training process is performed.

[11] The inference device described in [8], wherein the acquisition unit performs predetermined preprocessing on the newly acquired input data, and the trained model infers chemical structure information of lipid molecules associated with the preprocessed input data.

[12] An acquisition step of acquiring input data including prerequisites for designing or selecting lipid molecules constituting particles containing an active substance; An inference method comprising: an execution step of executing a learned model generated by performing a learning process on a learning model that associates input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules; The execution step is an inference method in which the chemical structure information of lipid molecules associated with the input data newly acquired in the acquisition step is inferred by executing the trained model.

[13] An acquisition step of acquiring input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; An inference program for causing a computer to execute the following steps: an execution step of executing a learned model generated by performing a learning process on a learning model that associates input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules; The execution step is an inference program that executes the trained model to infer chemical structure information of lipid molecules associated with the input data newly acquired in the acquisition step.

[14] A model generation method for generating a trained model by performing a learning process on a learning model that associates input data containing prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules.

[15] An acquisition unit that acquires from a user prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; A trained model generated by performing a training process on a training model that associates input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules; a providing unit that provides the user with chemical structure information of lipid molecules inferred by the trained model when input data including preconditions newly acquired by the user is input by the acquiring unit; and An inference service providing system having the above.

[16] The inference service providing system described in

[15] further includes a billing unit that charges the user when the learned model infers chemical structure information of lipid molecules by inputting input data including prerequisites newly acquired by the user by the acquisition unit.

[17]

[16] An inference service providing system as described in

[16] , wherein when the billing unit obtains from the user the transfection efficiency and / or cell survival rate of an active substance encapsulated in particles containing lipid molecules into cells, the transfection efficiency and / or cell survival rate being calculated from measurement results obtained by introducing the active substance encapsulated in particles containing lipid molecules having chemical structure information inferred by the trained model into cells, the billing unit changes the billing details for the user.

[18] An acquisition step of acquiring from a user prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; An execution step of executing a trained model generated by performing a training process on a training model that associates input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules; a providing step of providing the user with chemical structure information of lipid molecules inferred by the trained model by inputting input data including preconditions newly acquired by the user in the acquisition step; An inference service providing method having the steps of:

[19] An acquisition step of acquiring from a user prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; An execution step of executing a trained model generated by performing a training process on a training model that associates input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules; a providing step of providing the user with chemical structure information of lipid molecules inferred by the trained model by inputting input data including preconditions newly acquired by the user in the acquisition step; An inference service providing program for causing a computer to execute the above.

[20] an acquisition unit that acquires input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including the preconditions acquired by the acquisition unit; a calculation unit that calculates a reward based on the transfection efficiency of the active substance encapsulated in the particle containing the lipid molecule into the cell and / or the cell survival rate, the calculation being calculated from the measurement results obtained by introducing the active substance encapsulated in the particle containing the lipid molecule having the chemical structure information inferred by the reinforcement learning model into the cell, An inference device in which the reinforcement learning model performs learning processing based on the reward calculated by the calculation unit.

[21] The inference device according to

[20] , wherein the calculation unit calculates the reward so that it is maximized by increasing the transfection efficiency and / or cell survival rate.

[22] The inference device described in

[20] , wherein the acquisition unit performs predetermined preprocessing on the input data, and the reinforcement learning model infers chemical structure information of lipid molecules by inputting the preprocessed input data.

[23] An acquisition step of acquiring input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; an execution step of executing a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including the preconditions acquired in the acquisition step; and a calculation step of calculating a reward based on the transfection efficiency of the active substance encapsulated in the particles comprising lipid molecules into cells and / or the cell survival rate, the calculation being calculated from the measurement results obtained by introducing the active substance encapsulated in the particles comprising lipid molecules having chemical structure information inferred by the reinforcement learning model into cells, An inference method in which the reinforcement learning model performs a learning process based on the reward calculated in the calculation step.

[24] An acquisition step of acquiring input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; an execution step of executing a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including the preconditions acquired in the acquisition step; and a calculation step of calculating a reward based on the transfection efficiency of the active substance encapsulated in the particle comprising the lipid molecule into the cell and / or the cell survival rate, the calculation step being calculated from the measurement results obtained by introducing the active substance encapsulated in the particle comprising the lipid molecule having the chemical structure information inferred by the reinforcement learning model into the cell, An inference program in which the reinforcement learning model performs a learning process based on the reward calculated in the calculation step.

[25] An acquisition unit that acquires prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance from a user; a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including prerequisites acquired by a user via the acquisition unit; A providing unit that provides the user with chemical structure information of lipid molecules inferred by the reinforcement learning model; a calculation unit that calculates a reward based on the transfection efficiency of the active substance encapsulated in the particle containing the lipid molecule into the cell and / or the cell survival rate, the calculation being calculated from the measurement results obtained by introducing the active substance encapsulated in the particle containing the lipid molecule having the chemical structure information inferred by the reinforcement learning model into the cell, An inference service providing system in which the reinforcement learning model performs learning processing based on the reward calculated by the calculation unit.

[26] a billing unit that charges the user when the providing unit provides the user with the chemical structure information of the lipid molecule inferred by the reinforcement learning model; The inference service providing system according to

[25] , further comprising:

[27] An inference service providing system as described in

[26] , wherein when the billing unit obtains from the user the transfection efficiency and / or cell survival rate of an active substance encapsulated in a particle containing a lipid molecule, the transfection efficiency and / or cell survival rate of the active substance encapsulated in a particle containing a lipid molecule having chemical structure information inferred by the reinforcement learning model, calculated from measurement results obtained by introducing the active substance encapsulated in the particle into a cell, the billing unit changes the billing details for the user.

[28] An acquisition step of acquiring prerequisites for designing or selecting lipid molecules constituting particles containing an active substance from a user; an execution step of executing a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including the prerequisites acquired by the user in the acquisition step; a providing step of providing the user with chemical structure information of lipid molecules inferred by the reinforcement learning model; and a calculation step of calculating a reward based on the transfection efficiency of the active substance encapsulated in the particles containing the lipid molecules into cells and / or the cell survival rate, the calculation step being calculated from a measurement result obtained by introducing into cells an active substance encapsulated in the particles containing the lipid molecules having chemical structure information inferred by the reinforcement learning model, An inference service providing method in which the reinforcement learning model performs learning processing based on the reward calculated in the calculation step.

[29] An acquisition step of acquiring prerequisites for designing or selecting lipid molecules constituting particles containing an active substance from a user; an execution step of executing a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including the prerequisites acquired by the user in the acquisition step; a providing step of providing the user with chemical structure information of lipid molecules inferred by the reinforcement learning model; a calculation step of calculating a reward based on the transfection efficiency of an active substance encapsulated in a particle comprising a lipid molecule and / or a cell survival rate, the calculation step being calculated from a measurement result obtained by introducing an active substance encapsulated in a particle comprising a lipid molecule having chemical structure information inferred by the reinforcement learning model into a cell, An inference service providing program in which the reinforcement learning model performs learning processing based on the reward calculated in the calculation step.

[30] A trained model generated by performing a training process on a training model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of an active substance encapsulated in a particle containing the lipid molecules and / or the cell survival rate; a generation unit that, when the transfection efficiency and / or cell survival rate associated with input data including chemical structure information of a newly generated lipid molecule is inferred by the trained model, repeats a generation process of generating chemical structure information of a next new lipid molecule based on the inference result until a predetermined termination condition is satisfied; An inference device having:

[31] The generating unit: An inference device described in

[30] , which selects one of a plurality of search spaces corresponding to combinations of formable hydrocarbon molecular fragments and chemical skeletons of lipid molecules based on the inference results, and generates chemical structure information of the next new lipid molecule using the characteristics of the selected search space.

[32] The inference device described in

[31] , wherein the multiple search spaces have different combinations of the length, saturation, and number of branches of the molecular fragments and the type of chemical skeleton of the lipid molecules.

[33] The inference device described in

[31] , wherein the generation unit generates chemical structure information of the next new lipid molecule under predetermined constraints.

[34] Further comprising an acquisition unit for acquiring prerequisites for designing or selecting lipid molecules constituting particles containing an active substance; The inference device described in

[30] , wherein the generation unit generates chemical structure information of the next new lipid molecule using the acquired prerequisite conditions as the specified constraint conditions.

[35] An execution step of executing a trained model generated by performing a training process on a training model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of an active substance encapsulated in a particle containing the lipid molecules and / or the cell survival rate; a generation step in which, when the transfection efficiency and / or cell viability associated with input data including chemical structure information of a newly generated lipid molecule is inferred by the trained model, a generation process for generating chemical structure information of the next new lipid molecule based on the inference result is repeated until a predetermined termination condition is satisfied; A method of inference having the following structure:

[36] An execution step of executing a trained model generated by performing a training process on a training model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of an active substance encapsulated in a particle containing the lipid molecules and / or the cell survival rate; a generation step in which, when the transfection efficiency and / or cell viability associated with input data including chemical structure information of a newly generated lipid molecule is inferred by the trained model, a generation process for generating chemical structure information of the next new lipid molecule based on the inference result is repeated until a predetermined termination condition is satisfied; An inference program that allows a computer to execute the above. [Effects of the Invention]

[0008] The present disclosure can assist in the design or selection of chemical structures of lipid molecules that constitute particles that encapsulate active substances. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 shows an example of the accumulation of various data in the drug delivery system review process. [Figure 2] FIG. 2 is a diagram illustrating an application example of the inference device according to the first embodiment. [Figure 3] FIG. 3 is a diagram illustrating an example of a hardware configuration of the inference device. [Figure 4] FIG. 4 is a diagram illustrating an example of the functional configuration of the learning device according to the first embodiment. [Figure 5] FIG. 5 is a diagram illustrating an example of the functional configuration of the inference device according to the first embodiment. [Figure 6] FIG. 6 is an example of a flowchart showing the flow of a process for inferring transfection efficiency and / or cell viability. [Figure 7A] FIG. 7A is a diagram illustrating an embodiment of a learning device. [Figure 7B] FIG. 7B is a diagram illustrating an embodiment of an inference device. [Figure 8] FIG. 8 is a diagram illustrating an application example of the inference device according to the second embodiment. [Figure 9] FIG. 9 is a diagram illustrating an example of the functional configuration of a learning device according to the second embodiment. [Figure 10]FIG. 10 is a diagram illustrating an example of the functional configuration of the inference device according to the second embodiment. [Figure 11] FIG. 11 is an example of a flowchart showing the flow of a process for inferring chemical structure information of a lipid molecule. [Figure 12] FIG. 12 is a diagram illustrating an application example of the inference service providing system according to the third embodiment. [Figure 13] FIG. 13 is an example of a flowchart showing the flow of an inference service providing process. [Figure 14] FIG. 14 is a diagram illustrating an application example of the inference service providing system according to the fourth embodiment. [Figure 15] FIG. 15 is a diagram illustrating an example of the functional configuration of an inference device according to the fourth embodiment. [Figure 16] FIG. 16 is another example of a flowchart showing the flow of the inference service providing process. [Figure 17] FIG. 17 is a diagram illustrating an example of the functional configuration of an inference device according to the fifth embodiment. [Figure 18] FIG. 18 is an example of a flowchart showing the flow of the generation process. [Figure 19] FIG. 19 is a diagram illustrating an example of the functional configuration of an inference device according to the sixth embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, each embodiment will be described with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configurations are designated by the same reference numerals, and redundant description will be omitted.

[0011] [First embodiment] <Example of data accumulation in the drug delivery system review process> First, a general example of how various data are accumulated in a drug delivery system review process will be described. Figure 1 shows an example of how various data are accumulated in a drug delivery system review process.

[0012] As shown in FIG. 1 , in the drug delivery system investigation process 100, when designing (or selecting) the chemical structure of lipid molecules that constitute particles containing an active substance, first, the following prerequisites for design or selection (hereinafter simply referred to as “prerequisites for design”) are established: Attributes of the subject (a human or non-human animal (hereinafter, sometimes referred to as a "subject animal, etc.") 160 or an in vitro, in situ, or ex vivo cell and / or tissue 160), - 160 types of diseases of target animals, etc. Attributes of the included active substance (e.g., nucleic acid 140) (e.g., type of active substance (e.g., nucleic acid 140), chemical structure information, etc.), A target into which the particles of the complex 150 are introduced (specific cells in vivo or in vitro (in situ or ex vivo) of a target animal or the like 160), (See design prerequisites 101) are input to the designer 110. Note that particles containing an active substance are at least: The lipid molecules are mixed with the active substance (e.g., nucleic acid 140) to form complex particles; A case in which lipid molecules form an outer shell and an active substance (e.g., nucleic acid 140) is contained within the shell to form a complex particle. It is a concept that includes:

[0013] In one embodiment of the present disclosure, the lipid molecules used include, for example, cationic lipid molecules. Cationic lipids refer to lipids that have a net positive charge at a selected pH, such as physiological pH. Methods for producing particles containing lipid molecules and active substances include those described in, for example, WO 2016 / 021683, WO 2019 / 131839, and WO 2020 / 032184. In another embodiment of the present disclosure, the lipid molecules used include, for example, anionic lipid molecules, cholesterol derivative molecules, and amphiphilic lipid molecules.

[0014] The nucleic acid used in one embodiment of the present disclosure may be any molecule formed by polymerizing nucleotides and molecules functionally equivalent to the nucleotides. Examples include RNA, which is a polymer of ribonucleotides; DNA, which is a polymer of deoxyribonucleotides; a mixed polymer of ribonucleotides and deoxyribonucleotides; and a nucleotide polymer containing nucleotide analogs. Nucleic acids may also be nucleotide polymers containing nucleic acid derivatives. Nucleic acids may be single-stranded or double-stranded. Double-stranded nucleic acids also include double-stranded nucleic acids in which one strand hybridizes to the other strand under stringent conditions. The nucleic acid used in this embodiment is not particularly limited and may be, for example, a nucleic acid intended for the improvement of a disease, symptom, disorder, or pathological condition, or for the alleviation or prevention of the onset of a disease, symptom, disorder, or pathological condition (sometimes referred to herein as "disease treatment, etc."). It may also be a nucleic acid intended for regulating the expression of a desired protein that does not contribute to disease treatment, but is useful for research purposes. Specific examples of nucleic acids used in this embodiment include siRNA, miRNA, miRNAmimic, antisense nucleic acid, ribozyme, mRNA, decoy nucleic acid, aptamer, DNA, and analogs or derivatives thereof that have been artificially modified.

[0015] A designer 110 designs or selects a chemical structure of a lipid molecule based on his or her experience and know-how from design prerequisites 101. A lipid molecule 111 having a chemical structure designed or selected by the designer 110 is subjected to an experiment and evaluation process by an experimenter and evaluator 120, and evaluation data 121 is notified to the designer 110.

[0016] The designer 110 redoes the design or selection of the chemical structure of the lipid molecule based on the notified evaluation data 121. A lipid molecule 111′ (not shown) having a chemical structure redesigned or reselected by the designer 110 is subjected to experimentation and evaluation again by the experimenter and evaluator 120, and evaluation data 121′ (not shown) is notified to the designer 110.

[0017] The design or selection by the designer 110 and the experimental and evaluation processes by the experimenter and evaluator 120 are repeated multiple times. This allows the designer 110 to search for a more appropriate chemical structure of lipid molecules that constitute the particles that encapsulate the nucleic acid 140.

[0018] Subsequently, lipid molecules 130 are generated based on the chemical structure information 180 of the lipid molecules searched for by the designer 110, and nucleic acid 140 is encapsulated in particles containing the generated lipid molecules 130 to form particles of complex 150. Note that, in addition to nucleic acid 140, the particles of complex 150 may contain components other than lipid molecules and nucleic acids as needed. Examples of such components include appropriate amounts of stabilizers and antioxidants. These components may be pharmaceutically acceptable components.

[0019] The formed particles of the complex 150 are applied to a target animal, etc. 160, or to cells and / or tissues, etc. outside the body (in vitro, in situ, or ex vivo) 160. Changes caused by introducing the particles of the complex 150 into specific cells in the body (in vivo) of the target animal, etc. 160 (or changes caused by introducing the particles into specific cells 160 outside the body (in vitro, in situ, or ex vivo)) are measured using various measurement methods and measuring devices, and output as effect data 161.

[0020] The method for applying the particles of the complex 150 to the target animal 160 may be any method known to those skilled in the art. Specific examples of the method for application to the target animal 160 include intravenous, intramuscular, intraperitoneal, intracerebral (intracerebral) (intracerebral (intrabrospinal)), subcutaneous, intra-articular, intrasynovial, intrathecal, oral, topical, or inhalation administration, either as a bolus or by continuous infusion over a certain period of time. Furthermore, those skilled in the art can appropriately determine the number of administrations, dosage, and administration interval. Specific examples of the method for application to cells and / or tissues outside the body (in vitro, in situ, or ex vivo) include adding particles of the complex 150 to a container in which target cells are cultured and culturing for a certain period of time. Those skilled in the art can appropriately determine the number of additions, dosage, addition interval, culture conditions, culture period, and the like.

[0021] The effect data 161 includes the transfection efficiency of the nucleic acid 140 encapsulated in the particles containing the lipid molecules 130 into cells and / or the cell survival rate, calculated from the measurement results.

[0022] The transfection efficiency can be appropriately evaluated using known methods based on the attributes of the nucleic acid, which is the active substance. For example, when siRNA is used as the nucleic acid, it can be evaluated based on the knockdown rate of the expression of the gene targeted by the siRNA. More specifically, the expression level of a gene (e.g., mRNA) in a group administered with particles containing the siRNA (administration group) can be compared with the expression level of the gene in a control group (e.g., a group administered with nothing, a group administered with particles containing no siRNA, or a group administered with only a substance containing neither siRNA nor lipid molecules (e.g., physiological saline)), and the ratio of the gene expression level in the administration group to the gene expression level in the control group can be calculated. When siRNA is used, the lower the ratio, the higher the transfection efficiency can be determined. Furthermore, when mRNA is used as the nucleic acid, it can be evaluated based on the expression level of the protein encoded by the mRNA. More specifically, it can be evaluated by comparing the expression levels of the protein in the same manner as when siRNA is used. When mRNA is used, the higher the ratio, the higher the transfection efficiency can be determined. Those skilled in the art can appropriately select the method and instrument for measuring gene expression level, protein expression level, etc.

[0023] The cell viability after transfection can also be evaluated appropriately using known methods. For example, the cell number before application can be compared with the cell number after application, and the ratio of the cell number after application to the cell number before application can be measured. The method and device for measuring the cell number can be appropriately selected by those skilled in the art.

[0024] Examples of evaluation data and effect data include cell transfection efficiency and cell survival rate, but may also include, but are not limited to, data on the in vivo kinetics (absorption, distribution, metabolism) and toxicity of nucleic acid 140 when particles of complex 150 are applied to a target animal, etc. 160.

[0025] The various data acquired during this series of steps in the drug delivery system review process 100 are stored in a drug delivery system-related data storage unit 170. As shown in Figure 1, the various data stored in the drug delivery system-related data storage unit 170 include, for example, · Design Prerequisites 101, ·Evaluation data 121, Chemical structure information of 130 lipid molecules, 140 types of nucleic acids, chemical structure information, Chemical structure information for 150 complexes, efficacy data161 (including transfection efficiency and / or cell viability); etc. are included.

[0026] The chemical structure information of the lipid molecule is not particularly limited, but examples thereof include the chemical formula, three-dimensional structure, molecular weight, number of carbon atoms, number of nitrogen atoms, number of oxygen atoms, and charge.

[0027] The chemical structure information of a nucleic acid is not particularly limited, but examples thereof include the number of bases constituting the nucleic acid, chemical formula, three-dimensional structure, molecular weight, and charge.

[0028] Examples of the chemical structure information of the complex include the particle size and membrane potential of the complex.

[0029] These chemical structure information can be appropriately measured using known methods.

[0030] The various data stored in the data storage unit 170 relating to the drug delivery system may further include various publicly available information (for example, patent publications, papers) and data available from databases.

[0031] <Examples of application of inference devices> Next, we will explain an example of application of an inference device to a drug delivery system review process when the inference device is generated using various data stored in the data storage unit 170 related to the drug delivery system shown in Figure 1. Figure 2 is a diagram showing an example of application of the inference device according to the first embodiment.

[0032] As in FIG. 1 , in the drug delivery system consideration process 200 to which the inference device 220 according to the first embodiment is applied, when designing (or selecting) the chemical structure of lipid molecules constituting particles containing an active substance, the following design preconditions 201 are set: Attributes of the subject (subject animal, etc. 260 or in vitro, in situ, or ex vivo cells and / or tissues 260); - 260 types of diseases of target animals, etc. Attributes of the included active substance (e.g., nucleic acid 240) (e.g., type of active substance (e.g., nucleic acid 240), chemical structure information, etc.), A subject into which the particles of the complex 250 are introduced (specific cells in vivo or in vitro (in situ or ex vivo) specific cells 260 of a subject animal or the like 260), The above is input to the designer 110.

[0033] The designer 110 designs or selects the chemical structure of a lipid molecule based on his or her experience and know-how from the design prerequisites 201. Chemical structure information of a lipid molecule 211 having the chemical structure designed or selected by the designer 110 is input to the inference device 220.

[0034] Here, the inference device 220 has a trained model generated by the learning device 210. The learning device 210 generates the trained model by performing a learning process on the trained model using a training dataset generated based on various data stored in the data storage unit 170 related to the drug delivery system.

[0035] The inference device 220 generates evaluation data 221 about the chemical structure information of the lipid molecule 211 using the trained model generated by the learning device 210. The evaluation data 221 generated by the inference device 220 is notified to the designer 110.

[0036] The design or selection by the designer 110 and the generation of evaluation data by the inference device 220 are repeated multiple times, allowing the designer 110 to search for more appropriate chemical structures of lipid molecules that make up the particles that encase nucleic acid 240.

[0037] Subsequently, lipid molecules 230 are generated based on the chemical structure information 280 of the lipid molecules searched for by the designer 110, and nucleic acid 240 is encapsulated in particles containing the generated lipid molecules 230 to form particles of complex 250. Note that, in addition to nucleic acid 240, the particles of complex 250 may contain components other than lipid molecules and nucleic acids as needed. Examples of such components include appropriate amounts of stabilizers and antioxidants. These components may be pharmaceutically acceptable components.

[0038] The formed particles of complex 250 are applied to a target animal, etc. 260, or to cells and / or tissues, etc. outside the body (in vitro, in situ, or ex vivo) 260. Changes caused by introducing particles of complex 250 into specific cells in the body (in vivo) of target animal, etc. 260 (or changes caused by introducing particles of complex 250 into specific cells outside the body (in vitro, in situ, or ex vivo) 260) are measured by various measurement methods and measuring devices, and output as effect data 261.

[0039] Methods for applying particles of the conjugate 250 to the target animal 260 may be known to those skilled in the art. Specific examples of methods for application to the target animal 260 include intravenous, intramuscular, intraperitoneal, intracerebral (intracerebral) (intracerebral (intrabrospinal)), subcutaneous, intra-articular, intrasynovial, intrathecal, oral, topical, or inhalation administration, either as a bolus or by continuous infusion over a certain period of time. Furthermore, those skilled in the art can appropriately determine and design the number of administrations, dosage, and administration interval. Specific examples of methods for application to cells and / or tissues outside the body (in vitro, in situ, or ex vivo) include adding particles of the conjugate 250 to a container in which target cells are cultured and culturing for a certain period of time. Those skilled in the art can appropriately determine the number of administrations, dosage, administration interval, culture conditions, culture period, and the like.

[0040] Furthermore, changes caused by the introduction of particles of complex 250 into specific cells in vivo in a target animal 260 can be detected by measuring the target animal 260 into which particles of complex 250 have been introduced, or a sample containing the specific cells collected from the target animal 260. The sample is not particularly limited as long as it contains the specific cells, and examples include body fluids such as whole blood, plasma, urine, serum, lymph, saliva, anal and vaginal secretions, sweat, and semen, as well as tissue samples and cells obtained from organ or tissue biopsies. Furthermore, the collected sample may be labeled by any known method for measurement.

[0041] Furthermore, examples of "target animals, etc." include, but are not limited to, humans, as well as animals such as mice, rats, guinea pigs, dogs, cats, rabbits, cows, horses, sheep, goats, and pigs. Target animals, etc. may be healthy humans or animals, or humans (patients) or animals suffering from some kind of disease.

[0042] Furthermore, any in vitro or in vivo measurement method known to those skilled in the art may be used. Specific examples include flow cytometry, immunological assays, mRNA transcript analysis, PCR, and hybridization. Other examples include sequencing, RFLP, Western blot, ELISA, radioimmunoassay, immunoprecipitation, FACS, HPLC, surface plasmon resonance, optical spectroscopy, and mass spectrometry. Hereinafter, examples of measurement results include cell transfection efficiency and cell viability, but the measurement items are not limited to these.

[0043] The effect data 261 includes, calculated from the measurement results, the transfection efficiency of the nucleic acid 240 encapsulated in the particles containing the lipid molecules 230 into cells and / or the cell survival rate. The effect data 261 may further include, but is not limited to, data on the pharmacokinetics (absorption, distribution, metabolism) and toxicity of the nucleic acid 240 in vivo when the particles of the complex 250 are applied to the target animal or the like 260.

[0044] In this way, whereas conventionally evaluation data 121 was generated by experiment processing and evaluation processing performed by an experimenter and evaluator 120, by applying the inference device 220, evaluation data 221 can be generated without performing the experiment processing and evaluation processing.

[0045] This reduces the time required to search for a more appropriate chemical structure of lipid molecules that constitute particles that encapsulate nucleic acid 240. In other words, the inference device 220 according to the first embodiment can assist in the task of designing or selecting the chemical structure of lipid molecules that constitute particles that encapsulate nucleic acid.

[0046] <Hardware configuration of the learning device and inference device> Next, we will explain the hardware configurations of the learning device 210 and the inference device 220. Since the learning device 210 and the inference device 220 have similar hardware configurations, we will explain the hardware configuration of the inference device 220 here.

[0047] Fig. 3 is a diagram showing an example of the hardware configuration of an inference device. As shown in Fig. 3, the inference device 220 has a processor 301, a memory 302, an auxiliary storage device 303, an I / F (Interface) device 304, a communication device 305, and a drive device 306. The hardware components of the inference device 220 are connected to each other via a bus 307.

[0048] The processor 301 has various arithmetic devices such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), etc. The processor 301 reads various programs installed in the auxiliary storage device 303 onto the memory 302 and executes them.

[0049] The memory 302 has a main storage device such as a ROM (Read Only Memory), a RAM (Random Access Memory), etc. The processor 301 and the memory 302 form a so-called computer, and the processor 301 executes various programs read onto the memory 302, thereby enabling the computer to realize various functions.

[0050] The auxiliary storage device 303 stores various programs and various data used when the processor 301 executes the various programs.

[0051] The I / F device 304 is a connection device that connects the operation device 310 and the display device 311 with the inference device 220. The I / F device 304 accepts various instructions for the inference device 220 via the operation device 310. The I / F device 304 also outputs the processing results of the inference device 220 via the display device 311.

[0052] The communication device 305 is a communication device for communicating with other devices via a network.

[0053] The drive device 306 is a device for loading a recording medium 312. The recording medium 312 here includes media that record information optically, electrically, or magnetically, such as a CD-ROM, a flexible disk, a magneto-optical disk, etc. The recording medium 312 may also include semiconductor memory that records information electrically, such as a ROM, a flash memory, etc.

[0054] The various programs to be installed in the auxiliary storage device 303 are installed, for example, by setting the distributed recording medium 312 in the drive device 306 and reading the various programs recorded on the recording medium 312 by the drive device 306. Alternatively, the various programs to be installed in the auxiliary storage device 303 may be installed by being downloaded from a network via the communication device 305.

[0055] <Functional configuration of the learning device> Next, the functional configuration of the learning device 210 will be described in detail. Fig. 4 is a diagram showing an example of the functional configuration of the learning device according to the first embodiment. In Fig. 4, a learning dataset 400 is an example of a learning dataset generated based on various data stored in the data storage unit 170 related to the drug delivery system. Note that, although examples of information items of the learning data (input data and correct answer data) are shown below, they are not limited to these, and some or all of them can be used as appropriate as learning data.

[0056] As shown in Figure 4, the learning dataset 400 has input data and correct answer data, and the input data includes information items such as "type of disease," "type of nucleic acid," "target of introduction," "chemical structure information of lipid molecules," and "attribute information of target animals, etc."

[0057] In the "disease type", for example, when the target animal etc. 160 is a patient (hereinafter sometimes referred to as the "target patient" in this specification), the type of the disease, "disease A1", is stored.

[0058] In "type of nucleic acid", for example, "nucleic acid X1", "nucleic acid X2", etc. are stored as types of nucleic acid corresponding to "disease A1".

[0059] "Target of introduction" stores information indicating the target cell into which the complex particle should be introduced for each "type of disease" and "type of nucleic acid." The example in Figure 4 shows that when the disease type is "Disease A1" and the nucleic acid type is "Nucleic acid X1," the complex particle is introduced into "Cell Y1," and when the disease type is "Disease A1" and the nucleic acid type is "Nucleic acid X2," the complex particle is introduced into "Cell Y2."

[0060] Examples of information indicating the target cells include, but are not limited to, the source of origin or the organ or tissue in which the cells reside in the body, and the type of cell (e.g., nerve cells, parenchymal cells, interstitial cells, etc.).

[0061] The "chemical structure information of lipid molecules" stores information indicating the chemical structures of more appropriate lipid molecules that have been previously designed or selected by the designer 110 based on the design prerequisites 101 such as "type of disease," "type of nucleic acid," and "target of introduction."

[0062] The "attribute information of the target animal, etc." stores attribute information of the target animal, etc. 160 (e.g., a target patient). The example of Figure 4 shows that a particle of a complex formed by encapsulating a nucleic acid type = "nucleic acid X1" in a particle containing lipid molecules has been introduced into a "cell Y1" of a target patient having attribute information of the target animal, etc. = "patient attribute Z1." The example of Figure 4 also shows that a particle of a complex formed by encapsulating a nucleic acid type = "nucleic acid X2" in a particle containing lipid molecules has been introduced into a "cell Y2" of a target patient having attribute information of the target animal, etc. = "patient attribute Z2."

[0063] On the other hand, the correct answer data includes "transfection efficiency and / or cell survival rate" as an information item.

[0064] The "transfection efficiency and / or cell viability" field stores the transfection efficiency and / or cell viability of a nucleic acid encapsulated in a particle containing a lipid molecule. Specifically, for example, the transfection efficiency and / or cell viability calculated from the measurement results obtained by applying particles of the corresponding complex to a corresponding target animal 260 or cells and / or tissues 260 outside the body (in vitro, in situ, or ex vivo), and measuring the target animal or a sample isolated therefrom, is stored.

[0065] These training datasets may be data obtained from various publicly available information (for example, patent publications, papers) or databases.

[0066] The example in Figure 4 shows that the transfection efficiency of cells with nucleic acid type = "Nucleic acid X1" was "85%." The example in Figure 4 also shows that the transfection efficiency of cells with nucleic acid type = "Nucleic acid X2" was "72%."

[0067] On the other hand, a learning program is installed in the learning device 210, and by executing the learning program, the learning device 210 functions as a preprocessing unit 410, a learning model 420, and a comparison / modification unit 430.

[0068] The preprocessing unit 410 acquires the "input data" of the training dataset 400 and performs various preprocessing processes to generate preprocessed data suitable for input to the training model 420. The various preprocessing processes performed by the preprocessing unit 410 include normalizing the input data, vectorizing the input data, and the like.

[0069] The learning model 420 is a model that associates input data with correct data (transfection efficiency and / or cell survival rate). Specifically, the learning model 420 receives the preprocessed data notified by the preprocessing unit 410 as input and outputs the transfection efficiency and / or cell survival rate.

[0070] The learning model 420 is subjected to a learning process in which the model parameters are updated by backpropagating the error calculated by the comparison / change unit 430. This generates a trained model. That is, the trained model is generated by updating the model parameters of the learning model 420 so that the output of the learning model 420 approaches the correct data (transfection efficiency and / or cell survival rate).

[0071] The comparison / modification unit 430 calculates an error by comparing the transfection efficiency and / or cell survival rate output from the learning model 420 with the correct data (transfection efficiency and / or cell survival rate) of the learning dataset 400. The comparison / modification unit 430 also updates the model parameters of the learning model 420 by backpropagating the calculated error.

[0072] <Functional configuration of the inference device> Next, the functional configuration of the inference device 220 will be described in detail. Fig. 5 is a diagram showing an example of the functional configuration of the inference device according to the first embodiment. As shown in Fig. 5, the input data 500_1, 500_2, 500_3, ... includes Information included in the design prerequisites 201 (type of disease, type of nucleic acid, target of introduction, attribute information of target animal, etc.), Chemical structure information of lipid molecules designed or selected by the designer 110 based on the design prerequisites 201; Includes:

[0073] The input data may be data obtained from various publicly available information (for example, patent publications, papers) or databases.

[0074] Input data 500_1, 500_2, 500_3, ... containing chemical structure information of different lipid molecules is input to inference device 220. An inference program is installed in inference device 220, and by executing the inference program, inference device 220 functions as preprocessing unit 510, trained model 520, and evaluation data generation unit 530.

[0075] The preprocessing unit 510 is an example of an acquisition unit, and has the same functions as the preprocessing unit 410 of the learning device 210. Specifically, the preprocessing unit 510 acquires input data 500_1, 500_2, 500_3, and performs various preprocessing processes on the acquired input data 500_1, 500_2, 500_3, to generate preprocessed data.

[0076] The trained model 520 is a trained model generated by the learning device 210 through a learning process, and uses the preprocessed data notified by the preprocessing unit 510 as input to infer the transfection efficiency and / or cell survival rate.

[0077] The evaluation data generation unit 530 generates evaluation data 540_1, 540_2, 540_3, . . . based on the transfection efficiency and / or cell survival rate inferred by the trained model 520.

[0078] In this way, the inference device 220 can infer the transfection efficiency and / or cell viability of lipid molecules having chemical structures designed or selected by the designer 110 and generate evaluation data without performing experimental and evaluation processes.

[0079] This reduces the time required to search for a more appropriate chemical structure of lipid molecules that constitute particles that encapsulate nucleic acid 240. In other words, the inference device 220 according to the first embodiment can assist in the task of designing or selecting the chemical structure of lipid molecules that constitute particles that encapsulate nucleic acid.

[0080] <Process flow for inferring transfection efficiency and / or cell viability> Next, we will explain the flow of the transfection efficiency and / or cell viability inference process in the drug delivery system review process 200. Figure 6 is an example of a flowchart showing the flow of the transfection efficiency and / or cell viability inference process.

[0081] In step S601, the learning device 210 acquires various data related to the drug delivery system from the data storage unit 170.

[0082] In step S602, the learning device 210 generates the learning dataset 400 using the various acquired data.

[0083] In step S603, the learning device 210 performs a learning process on the learning model 420 using the learning dataset 400 to generate a trained model 520.

[0084] In step S604, the inference apparatus 220 acquires input data (for example, input data 500_1) including chemical structure information of a lipid molecule newly designed or selected by the designer 110.

[0085] In step S605, the inference device 220 executes the trained model 520 by inputting the acquired input data (e.g., input data 500_1) into the trained model 520, and infers the transfection efficiency and / or cell survival rate. The inference device 220 also generates evaluation data (e.g., evaluation data 540_1) based on the inferred transfection efficiency and / or cell survival rate.

[0086] In step S606, inference device 220 determines whether or not there is next input data (for example, input data 500_2, 500_3, . . . ).

[0087] If it is determined in step S606 that there is next input data (YES in step S606), the process returns to step S604.

[0088] On the other hand, if it is determined in step S606 that there is no next input data (NO in step S606), the transfection efficiency and / or cell survival rate inference process is terminated.

[0089] <Summary> As is clear from the above description, the inference device 220 according to the first embodiment: Obtain input data that includes at least the chemical structure information of lipid molecules. The device has a trained model that is generated by performing a training process on a training model that associates input data with the transfection efficiency of nucleic acids encapsulated in particles containing lipid molecules into cells and / or cell survival rate. The trained model infers transfection efficiency and / or cell viability associated with newly acquired input data.

[0090] As a result, the inference device 220 according to the first embodiment can shorten the time required to search for a more appropriate chemical structure of lipid molecules that constitute particles that encapsulate nucleic acids. In other words, the inference device 220 according to the first embodiment can support the work of designing or selecting the chemical structure of lipid molecules that constitute particles that encapsulate nucleic acids.

[0091] <<Example 1>> Below, we will explain specific examples of the learning device 210 and the inference device 220 according to the first embodiment. Note that the following examples are merely examples, and the learning device 210 and the inference device 220 according to the first embodiment are not limited to the following examples.

[0092] <Example of learning device> First, an example of the learning device 210 will be described. In this example, a training dataset is divided into a training dataset and a validation dataset, and a learning process is performed on a training model using the training dataset to generate a trained model. In this example, the inference accuracy of the inference device 220 was evaluated using the validation dataset.

[0093] FIG. 7A is a diagram showing one embodiment of a learning device. As shown in FIG. 7A, the preprocessing unit 410 has a conversion unit 710, which reads input data from a dataset for learning and converts it into molecular descriptors. The example of FIG. 7A shows how the chemical structure information of 75 lipid molecules, from the chemical structure information of lipid molecules 0001 to 0200, is read and converted into 200 molecular descriptors. In the example of FIG. 7A, reference numeral 701 indicates the chemical structure information of one lipid molecule among the chemical structure information of the 75 lipid molecules read by the conversion unit 710.

[0094] As shown in FIG. 7A, the preprocessing unit 410 includes an extraction unit 720, which removes molecular descriptors that are not suitable for use in the learning process from the 200 molecular descriptors. Specifically, the extraction unit 720 removes molecular descriptors with small variance from the 200 molecular descriptors, and also removes molecular descriptors that are found to be collinear. In this way, the extraction unit 720 extracts molecular descriptors that are suitable for use in the learning process. In FIG. 7A, reference numeral 711 denotes an example of a molecular descriptor extracted by the extraction unit 720, which is input to the learning model 420 as preprocessed data.

[0095] As shown in FIG. 7A, in this embodiment, the learning model 420 uses "Gradient Boosting Decision Tree" as a machine learning algorithm, and hyperparameters are optimized by K-fold cross validation.

[0096] 7A, in this embodiment, the learning model 420 is trained using the transfection efficiency as the correct answer data, and the learning model 420 is trained to associate the preprocessed data (molecular descriptors) with the transfection efficiency, generating a trained model 520 (see FIG. 7B).

[0097] <Example of an inference device> Next, an embodiment of the inference device 220 will be described. FIG. 7B is a diagram illustrating an embodiment of the inference device. As shown in FIG. 7B, the preprocessing unit 510 has a conversion unit 710, which reads input data from a dataset for verification and converts it into molecular descriptors. The example of FIG. 7B shows how the conversion unit 710 reads out chemical structure information for 16 lipid molecules, from lipid molecule chemical structure information 0201 to lipid molecule chemical structure information 0216, and converts it into 200 molecular descriptors. In the example of FIG. 7B, reference numeral 801 indicates the chemical structure information for one lipid molecule among the chemical structure information for the 16 lipid molecules read by the conversion unit 710.

[0098] Also, as shown in FIG. 7B, the molecular descriptors converted by the conversion unit 710 are input as preprocessed data to the trained model 520, and the trained model 520 infers the transfection efficiency.

[0099] In Figure 7B, reference numeral 802 indicates the results of evaluating the inference accuracy of the trained model 520 for the chemical structure information of 16 lipid molecules included in the validation dataset. Specifically, reference numeral 802 indicates the evaluation of the inference accuracy by comparing the inferred value and the actual measured value of transfection efficiency and using the correlation coefficient and mean absolute error as indicators. As shown by reference numeral 802, in this example, it was confirmed that the trained model 520 can infer the transfection efficiency of the chemical structure information of 16 lipid molecules with high inference accuracy.

[0100] [Second embodiment] Next, a second embodiment will be described. In the first embodiment, the learning device 210 performs the following steps based on various data stored in the data storage unit 170 related to the drug delivery system: Input data including chemical structure information of lipid molecules; Transfection efficiency and / or cell viability, We explained the case of generating a trained model that associates

[0101] In contrast to this, in the second embodiment, the learning device performs the following based on various data stored in the data storage unit 170 related to the drug delivery system: · Information contained in the design assumptions; - Chemical structure information of lipid molecules and Hereinafter, the second embodiment will be described, focusing on the differences from the first embodiment.

[0102] <Examples of application of inference devices> First, an example of application of the inference device according to the second embodiment to a drug delivery system will be described. Fig. 8 is a diagram showing an example of application of the inference device according to the second embodiment.

[0103] As shown in Figure 8, in the drug delivery system consideration process 800, in order to infer chemical structure information of more appropriate lipid molecules that constitute particles containing nucleic acids 240, design preconditions 201 (specifically, input data including information contained in design preconditions 201) are input into an inference device 820.

[0104] Here, the inference device 820 has a trained model generated by the learning device 810. The learning device 810 generates the trained model by performing a learning process on the trained model using a training dataset generated based on various data stored in the data storage unit 170 related to the drug delivery system.

[0105] The inference device 820 executes the trained model generated by the learning device 810, and infers the chemical structure information 280 of the lipid molecule from the information contained in the design preconditions 201 (including input data).

[0106] Next, in the drug delivery system consideration process 800, lipid molecules 230 are generated based on the chemical structure information 280 inferred by the inference device 820, and particles containing the generated lipid molecules 230 are encapsulated with nucleic acids 240 to form particles of complexes 250.

[0107] The formed particles of the complex 250 are applied to a target animal, etc. 260, or to cells and / or tissues, etc., outside the body (in vitro, in situ, or ex vivo). Changes caused by introducing the particles of the complex 250 into specific cells in the body (in vivo) of the target animal, etc. 260 (or changes caused by introducing the particles into specific cells 260 outside the body (in vitro, in situ, or ex vivo)) are measured by various measurement methods or measuring devices, and output as effect data 261. The method of applying the particles of the complex 250 to the target animal, etc. 260, or cells and / or tissues outside the body (in vitro, in situ, or ex vivo) can be the same as that of the above-mentioned [First Embodiment].

[0108] The effect data 261 includes the transfection efficiency and / or cell survival rate of the nucleic acid 240 encapsulated in the particles containing the lipid molecules 230 into the cells, calculated from the measurement results. The transfection efficiency and cell survival rate can be evaluated in the same manner as in the above-mentioned [First Embodiment].

[0109] Examples of effect data 261 include the transfection efficiency into cells and cell survival rate, but may also include, but are not limited to, data on the in vivo kinetics (absorption, distribution, metabolism) and toxicity of nucleic acid 240 when particles of complex 250 are applied to a target animal or the like 260.

[0110] In this way, while the design or selection of the chemical structure of a lipid molecule conventionally depended on the experience and know-how of the designer 110, the inference device 820 can directly infer the chemical structure information 280 of the lipid molecule from the information contained in the design prerequisites 201.

[0111] This makes it possible to design or select a more appropriate chemical structure of lipid molecules that constitute particles that encapsulate nucleic acid 240, without relying on the experience or know-how of designer 110. In other words, the inference device 820 according to the second embodiment can assist in the task of designing or selecting the chemical structure of lipid molecules that constitute particles that encapsulate nucleic acid.

[0112] <Functional configuration of the learning device> Next, the functional configuration of the learning device 810 will be described in detail. Fig. 9 is a diagram showing an example of the functional configuration of the learning device according to the second embodiment. In Fig. 9, a learning dataset 900 is an example of a learning dataset generated based on various data stored in the data storage unit 170 related to the drug delivery system. Examples of information items of the learning data (input data and correct answer data) are shown below, but are not limited to these, and some or all of these can be used as appropriate as learning data.

[0113] As shown in Figure 9, the training dataset 900 has input data and correct answer data, and the input data includes the following information items: "type of disease," "type of nucleic acid," "target of introduction," "transfection efficiency and / or cell survival rate," and "attribute information of target animal, etc."

[0114] In the "disease type", for example, if the target animal etc. 160 is a patient, the type of the disease, "disease A1", is stored.

[0115] In "type of nucleic acid", for example, "nucleic acid X1", "nucleic acid X2", etc. are stored as types of nucleic acid corresponding to "disease A1".

[0116] "Target of introduction" stores information indicating the target cell into which the complex particle should be introduced for each "type of disease" and "type of nucleic acid." The example in Figure 9 shows that when the disease type is "Disease A1" and the nucleic acid type is "Nucleic acid X1," the complex particle is introduced into "Cell Y1," and when the disease type is "Disease A1" and the nucleic acid type is "Nucleic acid X2," the complex particle is introduced into "Cell Y2."

[0117] Examples of information indicating the target cells include, but are not limited to, the source of origin or the organ or tissue in which the cells reside in the body, and the type of cell (e.g., nerve cells, parenchymal cells, interstitial cells, etc.).

[0118] The "transfection efficiency and / or cell viability" field stores the transfection efficiency and / or cell viability of a nucleic acid encapsulated in a particle containing a lipid molecule. Specifically, the transfection efficiency and / or cell viability are calculated from the measurement results obtained by applying particles of the corresponding complex to a corresponding target animal 160 or cells and / or tissues 160 outside the body (in vitro, in situ, or ex vivo), and measuring the target animal or a sample isolated therefrom.

[0119] These training datasets may be data obtained from various publicly available information (for example, patent publications, papers) or databases.

[0120] The example in Figure 9 shows that the transfection efficiency of the type of nucleic acid encapsulated in the particles containing lipid molecules = "Nucleic acid X1" into cells was "85%." Also, the example in Figure 8 shows that the transfection efficiency of the type of nucleic acid encapsulated in the particles containing lipid molecules = "Nucleic acid X2" into cells was "72%."

[0121] The "attribute information of the target animal, etc." stores attribute information of the target animal, etc. 160 (e.g., the target patient). The example of Figure 9 shows that a particle of a complex formed by encapsulating a nucleic acid type = "nucleic acid X1" in a particle containing lipid molecules has been introduced into a "cell Y1" of a target patient having attribute information of the target animal, etc. = "patient attribute Z1." The example of Figure 9 also shows that a particle of a complex formed by encapsulating a nucleic acid type = "nucleic acid X2" in a particle containing lipid molecules has been introduced into a "cell Y2" of a target patient having attribute information of the target animal, etc. = "patient attribute Z2."

[0122] On the other hand, the correct answer data includes, as an information item, "chemical structure information of lipid molecules." The "chemical structure information of lipid molecules" stores information indicating the chemical structure of more appropriate lipid molecules previously designed or selected by the designer 110 based on information included in the design prerequisites 101, such as "type of disease," "type of nucleic acid," and "target of introduction."

[0123] On the other hand, a learning program is installed in the learning device 810, and by executing the learning program, the learning device 810 functions as a preprocessing unit 910, a learning model 920, and a comparison / modification unit 930.

[0124] The preprocessing unit 910 acquires the "input data" of the training dataset 900 and performs various preprocessing processes to generate preprocessed data suitable for input to the training model 920. The various preprocessing processes performed by the preprocessing unit 910 include processes for normalizing the input data and vectorizing the input data.

[0125] The learning model 920 is a model that associates input data with correct data (chemical structure information of lipid molecules). Specifically, the learning model 920 receives the preprocessed data notified by the preprocessing unit 910 as input and outputs chemical structure information of lipid molecules.

[0126] Note that a learning process is performed on the learning model 920, in which the model parameters are updated by backpropagating the error calculated by the comparison / change unit 930. This generates a trained model. That is, the trained model is generated by updating the model parameters of the learning model 920 so that the output of the learning model 920 approaches the correct data (chemical structure information of lipid molecules).

[0127] The comparison / modification unit 930 calculates an error by comparing the chemical structure information of the lipid molecules output from the learning model 920 with the correct data (chemical structure information of the lipid molecules) in the learning dataset 900. The comparison / modification unit 930 also updates the model parameters of the learning model 920 by backpropagating the calculated error.

[0128] <Functional configuration of the inference device> Next, the functional configuration of the inference device 820 will be described in detail. Fig. 10 is a diagram showing an example of the functional configuration of the inference device according to the second embodiment. As shown in Fig. 10, the input data 1000 includes information included in the design prerequisites 201 (such as the type of disease, the type of nucleic acid, the target to be introduced, and attribute information of the target animal) and the target transfection efficiency and / or cell survival rate. The input data may be data obtained from various publicly available information (for example, patent publications and papers) or databases.

[0129] Input data 1000 is input to an inference device 820. An inference program is installed in the inference device 820, and by executing the inference program, the inference device 820 functions as a preprocessing unit 1010 and a trained model 1020.

[0130] The preprocessing unit 1010 is another example of an acquisition unit, and has the same functions as the preprocessing unit 910 of the learning device 810. Specifically, the preprocessing unit 1010 acquires input data 1000 and performs various preprocessing operations on the acquired input data 1000 to generate preprocessed data.

[0131] The trained model 1020 is a trained model generated by the learning device 810 through a learning process, and uses the preprocessed data notified by the preprocessing unit 1010 as input to infer chemical structure information 280 of lipid molecules.

[0132] In this way, the inference device 820 can directly infer chemical structure information of lipid molecules from information included in the design prerequisites. This allows a more appropriate chemical structure of lipid molecules that constitute particles containing nucleic acid 240 to be designed or selected without relying on the experience or know-how of designer 110. In other words, the inference device 820 according to the second embodiment can assist in the task of designing or selecting the chemical structure of lipid molecules that constitute particles containing nucleic acid.

[0133] <Flow of the process for inferring chemical structure information of lipid molecules> Next, we will explain the flow of the process of inferring chemical structure information of lipid molecules by the drug delivery system review process 800. Figure 11 is an example of a flowchart showing the flow of the process of inferring chemical structure information of lipid molecules.

[0134] In step S1101, the learning device 810 acquires various data related to the drug delivery system from the data storage unit 170.

[0135] In step S1102, the learning device 810 generates the learning dataset 900 using the various acquired data.

[0136] In step S1103, the learning device 810 performs a learning process on the learning model 920 using the learning dataset 900 to generate a trained model 1020.

[0137] In step S1104, the reasoning device 820 obtains input data (eg, input data 1000) including the information contained in the design preconditions 201 and the target transfection efficiency and / or cell viability.

[0138] In step S1105, the inference device 820 executes the trained model 1020 by inputting the acquired input data (e.g., input data 1000) into the trained model 1020, and infers chemical structure information of lipid molecules.

[0139] <Summary> As is clear from the above description, the inference device 820 according to the second embodiment: · Obtain input data including prerequisites for designing or selecting lipid molecules that constitute particles that encapsulate nucleic acids. The system has a trained model that is generated by performing a training process on a training model that associates input data including prerequisites for designing or selecting lipid molecules that make up particles containing nucleic acids with chemical structure information of the lipid molecules. The trained model infers the chemical structure information of lipid molecules associated with newly acquired input data.

[0140] As a result, the inference device 820 according to the second embodiment can design or select a more appropriate chemical structure of lipid molecules that constitute particles that encapsulate nucleic acids, without relying on the experience or know-how of a designer. In other words, the inference device 820 according to the second embodiment can assist in the task of designing or selecting the chemical structure of lipid molecules that constitute particles that encapsulate nucleic acids.

[0141] [Third embodiment] In the second embodiment, Generate a training dataset based on various data stored in the data storage unit 170 related to the drug delivery system; - A trained model is generated by performing a training process using the generated training dataset, By inputting input data, including design assumptions, into the generated trained model, chemical structure information of lipid molecules can be inferred. forming a complex particle by encapsulating the nucleic acid in a particle containing a lipid molecule generated based on the inferred chemical structure information; Applying the formed complex particles to cells and / or tissues of a target animal or in vitro (in vitro, in situ, or ex vivo); The case has been explained.

[0142] In contrast to this, in the third embodiment, Obtain design assumptions from users, By inputting input data containing the acquired design assumptions into the generated trained model, chemical structure information of lipid molecules is inferred. Providing the inferred chemical structure information to the user; The user generates lipid molecules based on the provided chemical structure information, and encapsulates the nucleic acid in particles containing the generated lipid molecules, thereby forming complex particles; Collecting data from users regarding the drug delivery system obtained by applying the formed particles of the complex to target animals or cells and / or tissues outside the body (in vitro, in situ, or ex vivo); Update the training dataset using the collected data on drug delivery systems, and then re-run the training process to update the trained model. The following describes the case.

[0143] As a result, according to the third embodiment, it becomes possible to search for more appropriate lipid molecules that constitute particles that encapsulate each of the various types of nucleic acids possessed by the user, and it is possible to accumulate a large amount of know-how for searching for more appropriate lipid molecules that constitute particles that encapsulate nucleic acids. The third embodiment will be described below, focusing on the differences from the first and second embodiments.

[0144] <Application example of inference service provision system> First, an example of application of an inference service providing system according to a third embodiment, which provides users with chemical structure information of lipid molecules, to a drug delivery system review process will be described. Fig. 12 is a diagram showing an example of application of the inference service providing system according to the third embodiment.

[0145] Specifically, Figure 12 shows a case in which the inference service providing system 1210 provides chemical structure information of lipid molecules to each user in response to requests from user 1220 (user name = "User 1"), user 1230 (user name = "User 2"), etc.

[0146] As shown in FIG. 12, the inference service providing system 1210 includes an inference device 820 and an information providing device 1211 .

[0147] Among these, the inference device 820 is the same as the inference device 820 described in the second embodiment above using Figures 8 and 10. Specifically, the inference device 820 executes the trained model generated by the learning device 810, and infers chemical structure information of lipid molecules from input data including design prerequisites.

[0148] On the other hand, the information providing device 1211 functions as an acquiring unit. Specifically, the information providing device 1211 acquires a design precondition 1221 (information name="design precondition 1") and a design precondition 1231 (information name="design precondition 2") from the user 1220 and the user 1230, respectively.

[0149] Furthermore, the information providing device 1211 generates input data including the acquired design preconditions 1221 and 1231, and notifies the inference device 820. As a result, the inference device 820 executes the trained model and infers chemical structure information of lipid molecules, as in the second embodiment.

[0150] Furthermore, the information providing device 1211 functions as a providing unit. Specifically, in response to notification of input data including a design precondition 1221, the information providing device 1211 provides the user 1220 with chemical structure information 1212_1 (information name="chemical structure information 1 of lipid molecule") inferred by the inference device 820. In addition, in response to notification of input data including a design precondition 1231, the information providing device 1211 provides the user 1230 with chemical structure information 1212_2 (information name="chemical structure information 2 of lipid molecule") inferred by the inference device 820.

[0151] Furthermore, the information providing device 1211 also functions as a billing unit, and charges each user when providing the chemical structure information of lipid molecules to each user. This allows the inference service providing system 1210 to receive compensation according to the inference service of the chemical structure information of lipid molecules. Note that charging refers to the process of recording the amount each user should pay to the inference service providing system 1210.

[0152] Meanwhile, the user 1220 transmits the attribute information of the target animal, etc. 1225, the type of disease of the target animal, etc. 1225, the attributes of the nucleic acid 1223, the design prerequisites 1221 for the target into which the particles of the complex 1224 are to be introduced, etc. to the inference service providing system 1210 via a terminal not shown.

[0153] Furthermore, in exchange for payment of a fee, the user 1220 is provided with chemical structure information 1212_1 of lipid molecules corresponding to the design preconditions 1221 from the inference service providing system 1210 via a terminal (not shown).

[0154] The user 1220 also generates a lipid molecule 1222 (type of lipid molecule = "lipid molecule 1") based on the provided lipid molecule chemical structure information 1212_1. The user 1220 also forms a particle of complex 1224 (type of complex = "complex 1") by incorporating a nucleic acid 1223 (type of nucleic acid = "nucleic acid 1") into a particle containing the generated lipid molecule 1222.

[0155] The user 1220 also applies the particles of the formed complex 1224 to a target animal, etc. 1225 or to cells and / or tissues, etc. outside the body (in vitro, in situ, or ex vivo) 1225. The changes that occur when the particles of the complex 1224 are introduced into specific cells of the target animal, etc. 1225 (or changes that occur when the particles are introduced into specific cells, etc. outside the body (in vitro, in situ, or ex vivo) 1225) are measured using various measurement methods and measuring devices, and are output as effect data 1226 (data name = "effect data 1").

[0156] The effect data 1226 includes the transfection efficiency of the nucleic acid 1223 encapsulated in the particles containing the lipid molecules 1222 into cells and / or the cell survival rate, calculated from the measurement results.

[0157] Furthermore, the user 1220 registers various data acquired during the series of steps of the drug delivery system review process 1200 in the drug delivery system related data storage unit 170. As shown in FIG. 12, the drug delivery system related data 1227 (data name = "drug delivery system related data 1") registered in the drug delivery system related data storage unit 170 includes the following: · “Design Prerequisites 1” "Chemical structure information of lipid molecules 1" Chemical structure information of "Nucleic Acid 1" Chemical structure information of "Complex 1" "Effective Data 1", etc. are included.

[0158] The data 1227 regarding the drug delivery system registered by the user 1220 may be data obtained by the user 1220 from various publicly available information (eg, patent publications, papers) or databases.

[0159] In response to the user 1220 registering data 1227 related to the drug delivery system, the inference service providing system 1210 refunds a portion of the fee received for providing the chemical structure information 1212_1 of the lipid molecule to the user 1220. In other words, the inference service providing system 1210 changes the billing details for the user 1220.

[0160] Similarly, the user 1230 transmits attribute information of the target animal, etc. 1235, and if the target animal, etc. is a patient, the type of disease, attributes of the nucleic acid 1233 (e.g., the type of nucleic acid 1233, chemical structure information), and design prerequisites 1231 for the target into which particles of the complex 1234 are to be introduced to the inference service providing system 1210 via a terminal not shown.

[0161] Furthermore, in exchange for payment of a fee, the user 1230 is provided with chemical structure information 1212_2 of lipid molecules corresponding to the design preconditions 1231 by the inference service providing system 1210 via a terminal (not shown).

[0162] The user 1230 also generates a lipid molecule 1232 (type of lipid molecule = "lipid molecule 2") based on the provided lipid molecule chemical structure information 1212_2. The user 1230 also forms a particle of complex 1234 (type of complex = "complex 2") by incorporating a nucleic acid 1233 (type of nucleic acid = "nucleic acid 2") into a particle containing the generated lipid molecule 1232.

[0163] The user 1230 also applies the particles of the formed complex 1234 to a target animal, etc. 1235, or to cells and / or tissues, etc. outside the body (in vitro, in situ, or ex vivo) 1235. The changes that occur when the particles of the complex 1234 are introduced into specific cells in the body (in vivo) of the target animal, etc. 1235 (or the changes that occur when the particles are introduced into specific cells 1235 outside the body (in vitro, in situ, or ex vivo)) are measured using various measurement methods and measuring devices, and are output as effect data 1236 (data name = "effect data 2").

[0164] The effect data 1236 includes the transfection efficiency of the nucleic acid 1233 encapsulated in the particle comprising the lipid molecule 1232 into the cell and / or the cell survival rate, calculated from the measurement results.

[0165] Furthermore, the user 1230 registers various data acquired during the series of steps of the drug delivery system review process 1200 in the drug delivery system related data storage unit 170. As shown in FIG. 12, the drug delivery system related data 1237 (data name = "drug delivery system related data 2") registered in the drug delivery system related data storage unit 170 includes the following: · “Design Prerequisites 2” "Chemical structure information of lipid molecules 2" Chemical structure information of "Nucleic Acid 2" Chemical structure information of "Complex 2" "Effective Data 2", etc. are included.

[0166] In response to the user 1230 registering data 1237 related to the drug delivery system, the inference service providing system 1210 can refund a portion of the fee received for providing the chemical structure information 1212_2 of the lipid molecule to the user 1230. In other words, the inference service providing system 1210 can change the details of the fee charged to the user 1230.

[0167] In this way, the inference service providing system 1210 provides chemical structure information of lipid molecules in response to requests from each user, making it possible to search for more appropriate lipid molecules that make up particles that contain each of the various types of nucleic acids.

[0168] Furthermore, the data storage unit 170 relating to the drug delivery system can store data relating to the drug delivery system for many types of nucleic acids. The data stored in the data storage unit 170 relating to the drug delivery system may include various publicly available information (e.g., patent publications, papers) and data available from databases. Furthermore, the learning device 810 can update the training dataset using newly stored data relating to the drug delivery system and perform the training process again to update the trained model.

[0169] As a result, the inference service providing system 1210 according to the third embodiment can accumulate a large amount of know-how for searching for more appropriate lipid molecules that constitute particles that encapsulate nucleic acids. In other words, the inference service providing system 1210 according to the third embodiment can support the work of designing or selecting the chemical structures of lipid molecules that constitute particles that encapsulate nucleic acids.

[0170] <Inference service provision process flow> Next, a description will be given of the flow of the inference service provision process by the drug delivery system review process 1200. Fig. 13 is an example of a flowchart showing the flow of the inference service provision process.

[0171] In step S1301, the learning device 810 acquires various data related to the drug delivery system from the data storage unit 170.

[0172] In step S1302, the learning device 810 generates a learning dataset 900 using the various acquired data.

[0173] In step S1303, the learning device 810 performs a learning process on the learning model 920 using the learning dataset 900 to generate a trained model 1020.

[0174] In step S1304, the inference device 820 of the inference service providing system 1210 acquires the input data generated by the information providing device 1211 based on the preconditions for the design transmitted by the user.

[0175] In step S1305, the inference device 820 of the inference service providing system 1210 executes the trained model 1020 by inputting the acquired input data into the trained model 1020, and infers chemical structure information of lipid molecules.

[0176] In step S1306, the information providing device 1211 of the inference service providing system 1210 provides the chemical structure information of the lipid molecule inferred by the inference device 820 to the user who sent the design prerequisites, and charges the user.

[0177] In step S1307, if the inference service providing system 1210 collects data regarding a drug delivery system from the user in response to the information providing device 1211 providing the user with chemical structure information of lipid molecules, the inference service providing system 1210 can refund a portion of the fee to the user.

[0178] In step S1308, the inference service providing system 1210 determines whether a predetermined amount of data relating to the drug delivery system has been collected from the user.

[0179] If it is determined in step S1308 that a predetermined amount of data has been collected (YES in step S1308), the process returns to step S1302. In this case, a learning dataset is generated based on the newly registered predetermined amount of data, and the learning process is performed again.

[0180] On the other hand, if it is determined in step S1308 that the predetermined amount of data has not been collected (NO in step S1308), the process proceeds to step S1309.

[0181] In step S1309, the information providing device 1211 of the inference service providing system 1210 determines whether or not to end the inference service providing process. If it is determined in step 1309 that the inference service providing process is to be continued (NO in step S1309), the process returns to step S1304.

[0182] On the other hand, if it is determined in step S1309 that the inference service providing process is to be ended (YES in step S1309), the inference service providing process is ended.

[0183] <Summary> As is clear from the above description, the inference service providing system 1210 according to the third embodiment: · Obtain from the user prerequisites for designing or selecting lipid molecules that constitute particles that encapsulate nucleic acids. The system has a trained model that is generated by performing a training process on a training model that associates input data including prerequisites for designing or selecting lipid molecules that make up particles containing nucleic acids with chemical structure information of the lipid molecules. By inputting input data including preconditions obtained from the user, the chemical structure information of lipid molecules inferred by the trained model is provided to the user who submitted the preconditions.

[0184] As a result, the inference service providing system 1110 according to the third embodiment makes it possible to search for more appropriate lipid molecules that constitute particles that contain each of the many types of nucleic acids presented by the user.

[0185] Furthermore, the inference service providing system 1210 according to the third embodiment includes: In response to providing chemical structure information of lipid molecules, data on the drug delivery system is collected from the user.

[0186] As a result, the inference service providing system 1210 according to the third embodiment can accumulate a large amount of know-how for searching for more appropriate lipid molecules that constitute particles that encapsulate nucleic acids. In other words, the inference service providing system 1210 according to the third embodiment can support the work of designing or selecting the chemical structures of lipid molecules that constitute particles that encapsulate nucleic acids.

[0187] [Fourth embodiment] In the above third embodiment, it was explained that in response to providing chemical structure information of lipid molecules, data regarding the drug delivery system is collected from the user, and when a predetermined amount of data has been accumulated, the learning process is performed again.

[0188] In contrast, in the fourth embodiment, transfection efficiency and / or cell viability are obtained from the user in response to providing chemical structure information of lipid molecules. Furthermore, in the fourth embodiment, a reinforcement learning process is performed on a reinforcement learning model using a reward calculated based on the obtained transfection efficiency and / or cell viability. As a result, according to the fourth embodiment, it is possible to improve inference accuracy as the inference service is provided, and a large amount of know-how can be accumulated for searching for more appropriate lipid molecules that constitute particles containing nucleic acids.

[0189] The fourth embodiment will be described below, focusing on the differences from the third embodiment.

[0190] <Application example of inference service provision system> First, an example of application of an inference service providing system according to a fourth embodiment, which provides users with chemical structure information of lipid molecules, to a drug delivery system review process will be described. Fig. 14 is a diagram showing an example of application of the inference service providing system according to the fourth embodiment.

[0191] Specifically, Fig. 14 shows a case in which an inference service providing system 1410 provides chemical structure information of lipid molecules to each user in response to requests from user 1220, user 1230, etc., similar to Fig. 12. However, as shown in Fig. 14, in the case of the drug delivery system consideration process 1400, the inference service providing system 1410 has an inference device 1420 and an information providing device 1211.

[0192] Of these, the information providing device 1211 has the same functions as the information providing device 1211 described in the third embodiment using Fig. 12. Specifically, when design prerequisites are transmitted from each user, the information providing device 1211 generates input data and notifies the inference device 1420. In addition, in response to notifying the input data, the information providing device 1211 acquires chemical structure information of lipid molecules inferred by the inference device 1420 and provides the information to the corresponding user.

[0193] Furthermore, similar to the third embodiment, the information providing device 1211 charges each user for providing the chemical structure information of lipid molecules, thereby enabling the inference service providing system 1410 to receive payment in accordance with the inference service for the chemical structure information of lipid molecules.

[0194] Meanwhile, the inference device 1420 acquires effect data from the user in response to the chemical structure information of lipid molecules being provided to the user by the information providing device 1211. Furthermore, the inference device 1420 calculates a reward based on the transfection efficiency and / or cell survival rate included in the acquired effect data, and updates the model parameters of the reinforcement learning model based on the calculated reward.

[0195] In addition, the inference device 1420 executes the reinforcement learning model by inputting newly notified input data from the information providing device 1211 into the reinforcement learning model with updated model parameters, and infers new chemical structure information of lipid molecules.

[0196] 14, the processes performed by user 1220 and user 1230 are similar to those described in the third embodiment using FIG. 12, and therefore will not be described here. However, user 1220 transmits effect data 1226 to inference service providing system 1410. At this time, the transmitted effect data 1226 includes the transfection efficiency of nucleic acid 1223 encapsulated in particles containing lipid molecules 1222 into cells and / or the cell survival rate.

[0197] Similarly, the user 1230 transmits effect data 1236 to the inference service providing system 1410. At this time, the transmitted effect data 1236 includes the transfection efficiency of the nucleic acid 1233 encapsulated in the particle containing the lipid molecule 1232 into the cell and / or the cell survival rate.

[0198] The data sent to the inference service providing system may be data available from various publicly available information (for example, patent publications, papers) or databases.

[0199] In response to the user 1220 transmitting the effect data 1226, the inference service providing system 1410 can refund a portion of the fee for providing the lipid molecule chemical structure information 1212_1 to the user 1220. In other words, the inference service providing system 1410 can change the details of the fee charged to the user 1220.

[0200] Similarly, in response to the user 1230 sending the effect data 1236, the inference service providing system 1410 can refund part of the fee for providing the chemical structure information 1212_2 of the lipid molecule to the user 1230. In other words, the inference service providing system 1410 can change the details of the fee charged to the user 1230.

[0201] In this way, the inference service providing system 1410 performs reinforcement learning processing on the reinforcement learning model by obtaining effect data from each user each time it provides chemical structure information of lipid molecules in response to a request from each user.

[0202] As a result, according to the inference service providing system 1410 of the fourth embodiment, it becomes possible to perform reinforcement learning processing on a reinforcement learning model for a wide variety of nucleic acids, thereby improving inference accuracy as the inference service is provided.

[0203] As a result, the inference service providing system 1410 according to the fourth embodiment can accumulate a large amount of know-how for searching for more appropriate lipid molecules that constitute particles that encapsulate nucleic acids. In other words, the inference service providing system 1410 according to the fourth embodiment can support the work of designing or selecting more appropriate chemical structures of lipid molecules that constitute particles that encapsulate nucleic acids.

[0204] <Functional configuration of the inference device> Next, we will explain the details of the functional configuration of the inference device 1420. Fig. 15 is a diagram showing an example of the functional configuration of the inference device. An inference program is installed in the inference device 1420, and by executing the inference program, the inference device 1420 functions as a preprocessing unit 1510, a reinforcement learning model 1520, and a reward calculation unit 1530.

[0205] The preprocessing unit 1510 has the same functions as the preprocessing unit 910 of the learning device 810, and generates preprocessed data by performing various preprocessing operations on input data 1501, 1502, etc. The input data 1501 includes a design precondition 1221 sent by the user 1220. The input data 1502 includes a design precondition 1231 sent by the user 1230.

[0206] The reinforcement learning model 1520 receives the preprocessed data notified by the preprocessing unit 1510 as an input, and infers chemical structure information 1212_1, 1212_2, etc. of lipid molecules.

[0207] The reward calculation unit 1530 functions as a calculation unit and calculates a reward based on the transfection efficiency and / or cell survival rate included in the effect data 1226, 1236, etc. transmitted by the users 1220, 1230. The reward calculation unit 1530 calculates the reward so that the reward is maximized by increasing the transfection efficiency and / or cell survival rate.

[0208] Furthermore, the reward calculation unit 1530 performs a reinforcement learning process to update the model parameters of the reinforcement learning model 1520 based on the calculated reward.

[0209] <Inference service provision process flow> Next, a description will be given of the flow of the inference service provision process by the drug delivery system review process 1400. Fig. 16 is another example of a flowchart showing the flow of the inference service provision process.

[0210] In step S1601, the inference device 1420 of the inference service providing system 1410 acquires input data generated by the information providing device 1211 based on the preconditions for the design sent by the user.

[0211] In step S1602, the inference device 1420 of the inference service providing system 1410 inputs the acquired input data to the reinforcement learning model 1520, thereby executing the reinforcement learning model 1520 and inferring chemical structure information of lipid molecules.

[0212] In step S1603, the information providing device 1211 of the inference service providing system 1410 provides the chemical structure information of the lipid molecule inferred by the inference device 1420 to the user who sent the design prerequisites, and charges the user.

[0213] In step S1604, the inference device 1420 of the inference service providing system 1410 acquires effect data from the user in response to the provision of chemical structure information of lipid molecules to the user by the information providing device 1211. The inference device 1420 of the inference service providing system 1410 also refunds a portion of the fee to the user who sent the effect data.

[0214] In step S1605, the inference device 1420 of the inference service providing system 1410 calculates a reward based on the transfection efficiency and / or cell survival rate included in the acquired effect data.

[0215] In step S1606, the inference device 1420 of the inference service providing system 1410 performs a reinforcement learning process to update the model parameters of the reinforcement learning model based on the calculated reward.

[0216] In step S1607, the information providing device 1211 of the inference service providing system 1410 determines whether or not to end the inference service providing process. If it is determined in step 1607 that the inference service providing process is to be continued (NO in step S1607), the process returns to step S1601.

[0217] On the other hand, if it is determined in step S1607 that the inference service providing process is to be ended (YES in step S1607), the inference service providing process is ended.

[0218] <Summary> As is clear from the above description, the inference service providing system 1410 according to the fourth embodiment: · Obtain from the user prerequisites for designing or selecting lipid molecules that constitute particles that encapsulate nucleic acids. The system has a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including the acquired preconditions. The transfection efficiency and / or cell survival rate of the nucleic acid encapsulated in the particle containing the lipid molecule generated based on the chemical structure information inferred by the reinforcement learning model is obtained, and a reward is calculated. Based on the calculated reward, reinforcement learning processing is performed on the reinforcement learning model.

[0219] As a result, the inference service providing system 1410 according to the fourth embodiment can improve inference accuracy as the inference service is provided. As a result, the inference service providing system 1410 according to the fourth embodiment can accumulate a large amount of know-how for searching for more appropriate lipid molecules that constitute particles that encapsulate nucleic acids. In other words, the inference service providing system 1410 according to the fourth embodiment can support the work of designing or selecting more appropriate chemical structures of lipid molecules that constitute particles that encapsulate nucleic acids.

[0220] [Fifth embodiment] In the first embodiment, the case where the inference device 220 is configured to apply the trained model 520 generated by the learning device 210 to the inference device 220 and generate the evaluation data 221 has been described.

[0221] However, the method of applying the trained model 520 is not limited to this, and for example, the inference device may be configured to search for chemical structure information of lipid molecules that satisfy the target transfection efficiency and / or cell viability. The following describes the fifth embodiment, focusing on the differences from the first embodiment.

[0222] <Functional configuration of the inference device> First, the functional configuration of the inference device according to the fifth embodiment will be described in detail. Fig. 17 is a diagram showing an example of the functional configuration of the inference device according to the fifth embodiment. As shown in Fig. 17, the inference device 1700 functions as a preprocessing unit 510, a trained model 520, and a generation unit 1710.

[0223] Of these, the preprocessing unit 510 and the trained model 520 have already been described in the first embodiment using FIG. 5, and therefore a description thereof will be omitted here.

[0224] The generation unit 1710 has a reinforcement learning function using, for example, Thompson Sampling. Specifically, the generation unit 1710 determines whether a predetermined termination condition is met (for example, whether the transfection efficiency and / or cell survival rate inferred by the trained model 520 meets the target transfection efficiency and / or cell survival rate).

[0225] Furthermore, if the generation unit 1710 determines that the predetermined termination condition is not satisfied, it generates chemical structure information of the lipid molecules based on the transfection efficiency and / or cell survival rate inferred by the trained model 520. Furthermore, the generation unit 1710 notifies the preprocessing unit 510 of the generated chemical structure information of the lipid molecules.

[0226] On the other hand, if the generation unit 1710 determines that the specified termination condition is met, it outputs the chemical structure information of the lipid molecule generated last time as chemical structure information of the lipid molecule that satisfies the target transfection efficiency and / or cell survival rate.

[0227] In Fig. 17, reference numeral 1730 denotes an example of chemical structure information of a lipid molecule output from the generation unit 1710. The example in Fig. 17 shows how a specific lipid molecule is generated as chemical structure information of a lipid molecule that satisfies the target transfection efficiency and / or cell survival rate.

[0228] <Flow of generation process by inference device> Next, the flow of the generation process by the inference device 1700 for generating chemical structure information of lipid molecules that satisfy the target transfection efficiency and / or cell survival rate will be described.

[0229] Fig. 18 is an example of a flowchart showing the flow of the generation process. It is assumed that the target transfection efficiency and / or cell survival rate has been set in advance in inference device 1700 when inference device 1700 starts the generation process shown in Fig. 18.

[0230] In step S1801, the generating unit 1710 generates a group of molecular fragments from chemically formable hydrocarbons, with maximum values ​​set for length, degree of saturation, and number of branches.

[0231] In step S1802, the generation unit 1710 generates a group of chemical structures of lipid molecules by combining the generated molecular fragments with a group of chemical skeletons of lipids selected by the designer 110. At this time, the generation unit 1710 divides the group of chemical structures of the generated lipid molecules into a number of search spaces specified by the designer 110 according to the combination of the length, degree of saturation, number of branches, and type of chemical skeleton of the molecular fragments.

[0232] In step S1803, the generation unit 1710 uses Thompson Sampling to select one of the search spaces from the multiple search spaces divided in step S1802. Note that selecting one of the search spaces is nothing more than selecting the characteristics of the search space (combination of the length, saturation, number of branches, and type of chemical skeleton of the molecular fragment of the lipid molecule, etc.).

[0233] In step S1804, the generation unit 1710 generates a group of chemical structures of lipid molecules using a combination of the length, degree of saturation, number of branches, and type of chemical skeleton of the molecular fragment of the selected lipid molecule. Note that the generation unit 1710 randomly acquires multiple molecular fragments with a certain probability from a search space other than the selected search space, and generates a group of chemical structures of lipid molecules together with the selected molecular fragments.

[0234] In step S1805, the generation unit 1710 notifies the preprocessing unit 510 of each piece of chemical structure information of the generated chemical structure group of lipid molecules. The preprocessing unit 510 also performs various preprocessing operations on each piece of chemical structure information of the lipid molecules notified by the generation unit 1710 to generate a preprocessed data group suitable for input to the trained model 520, and inputs the preprocessed data group to the trained model 520. As a result, in step S1804, the trained model 520 infers the transfection efficiency and / or cell survival rate for the chemical structure group of lipid molecules generated by the generation unit 1710.

[0235] In step S1806, the generation unit 1710 The maximum value (inference result) of the multiple transfection efficiencies and / or cell viability rates inferred by the trained model 520; The search space selected in step S1803; The probability distribution used for Thompson Sampling is updated using

[0236] In step S1807, the generating unit 1710 determines whether or not a predetermined termination condition is satisfied. If it is determined in step S1807 that the predetermined termination condition is not satisfied (NO in step S1807), the process returns to step S1803.

[0237] On the other hand, if it is determined in step S1807 that the predetermined termination condition is met (YES in step S1807), the process proceeds to step S1808.

[0238] In step S1808, the generation unit 1710 outputs the chemical structure information of the lipid molecule generated last time (the chemical structure information of the lipid molecule whose maximum value was inferred) as the chemical structure information of the lipid molecule that satisfies the target transfection efficiency and / or cell survival rate.

[0239] <Summary> As is clear from the above description, the inference device 1700 according to the fifth embodiment: The system has a trained model that has undergone a training process to correlate input data including at least chemical structure information of lipid molecules with the transfection efficiency of an active substance encapsulated in a particle containing the lipid molecules and / or the cell survival rate. A generation unit is provided that, when the transfection efficiency and / or cell viability associated with input data including chemical structure information of a newly generated lipid molecule is inferred by the trained model, generates chemical structure information of the next new lipid molecule based on the inference result. The generation unit repeats the generation process of generating the next new chemical structure of a lipid molecule based on the inference result until a predetermined termination condition is met.

[0240] As a result, the inference device 1700 according to the fifth embodiment can generate, for example, chemical structure information of lipid molecules that satisfy a target transfection efficiency and / or cell survival rate. In other words, the inference device 1700 according to the fifth embodiment can assist in the task of designing or selecting chemical structure information of lipid molecules that constitute particles that encapsulate nucleic acids.

[0241] [Sixth embodiment] In the fifth embodiment, the chemical structure information of lipid molecules is generated according to the transfection efficiency and / or cell survival rate.

[0242] In contrast, in the sixth embodiment, similar to the second embodiment, in addition to the transfection efficiency and / or cell survival rate, information included in the design preconditions 201 (such as the type of disease, the type of nucleic acid, the target of introduction, and attribute information of the target animal) is input. Then, in the sixth embodiment, when generating chemical structure information of lipid molecules, the chemical structure information of lipid molecules is generated according to the information included in the design preconditions 201 (such as the type of disease, the type of nucleic acid, the target of introduction, and attribute information of the target animal). The sixth embodiment will be described below, focusing on the differences from the fifth embodiment.

[0243] <Functional configuration of the inference device> 19 is a diagram illustrating an example of the functional configuration of an inference device according to the sixth embodiment. As shown in FIG. 19, an inference device 1900 functions as a preprocessing unit 510, a trained model 520, a generation unit 1910, and an acquisition unit 1930.

[0244] Of these, the preprocessing unit 510 and the trained model 520 have already been described in the first embodiment using FIG. 5, and therefore a description thereof will be omitted here.

[0245] Like the generation unit 1710 in FIG. 17 , the generation unit 1910 has a reinforcement learning function using Thompson Sampling. Specifically, the generation unit 1910 acquires the transfection efficiency and / or cell survival rate output from the trained model 520. The generation unit 1910 also determines whether a predetermined termination condition is satisfied (e.g., whether the acquired transfection efficiency and / or cell survival rate satisfy the target transfection efficiency and / or cell survival rate). If the generation unit 1910 determines that the predetermined termination condition is not satisfied, the generation unit 1910 generates chemical structure information of lipid molecules based on the acquired transfection efficiency and / or cell survival rate, and notifies the pre-processing unit 510 of the generated chemical structure information of the lipid molecules.

[0246] When generating the chemical structure information of the lipid molecules, the generation unit 1910 acquires, as predetermined constraints, information included in the design prerequisites 201 (such as the type of disease, the type of nucleic acid, the target to be introduced, and attribute information of the target animal) notified by the acquisition unit 1930. When generating the chemical structure information of the lipid molecules, the generation unit 1910 generates the chemical structure information using the acquired information as predetermined constraints.

[0247] On the other hand, if the generation unit 1910 determines that the specified termination condition is met, it outputs the chemical structure information of the lipid molecule generated last time as chemical structure information of the lipid molecule that satisfies the target transfection efficiency and / or cell survival rate.

[0248] The acquisition unit 1930 acquires the input data 1000. The acquisition unit 1930 also notifies the generation unit 1910 of the acquired input data 1000.

[0249] <Summary> As is clear from the above description, the inference device 1900 according to the sixth embodiment: The system has a trained model that has undergone a training process to correlate input data including at least chemical structure information of lipid molecules with the transfection efficiency of an active substance encapsulated in a particle containing the lipid molecules and / or the cell survival rate. The trained model is used to infer transfection efficiency and / or cell viability associated with input data containing chemical structure information of newly generated lipid molecules. Based on the inferred transfection efficiency and / or cell viability, chemical structure information of the next new lipid molecule is generated, using the information included in the design prerequisites as a predetermined constraint. Repeat the generation process to generate chemical structure information for the next new lipid molecule based on the inferred transfection efficiency and / or cell viability and the information contained in the design assumptions, until a predetermined termination condition is met.

[0250] As a result, the inference device 1900 according to the sixth embodiment can generate chemical structure information of lipid molecules that satisfy the target transfection efficiency and / or cell survival rate under predetermined constraints. In other words, the inference device 1900 according to the sixth embodiment can assist in the design or selection of chemical structures of lipid molecules that constitute particles that encapsulate nucleic acids.

[0251] [Other embodiments] In the first to third embodiments described above, the learning device and the inference device are configured as separate devices. However, the learning device and the inference device may be configured as an integrated device.

[0252] Furthermore, in the above third embodiment, the inference device 820 and the information providing device 1211 are described as being configured as separate devices in the inference service providing system 1210. However, the inference device 820 and the information providing device 1211 may be configured as an integrated device. In this case, in the inference service providing system 1210, the functions of the inference device 820 and the functions of the information providing device 1211 may be realized by, for example, executing an inference service providing program in the integrated device.

[0253] Similarly, in the above fourth embodiment, the inference device 1420 and the information providing device 1211 are described as being configured as separate devices in the inference service providing system 1410. However, the inference device 1420 and the information providing device 1211 may be configured as an integrated device. In this case, in the inference service providing system 1410, the functions of the inference device 1420 and the functions of the information providing device 1211 may be realized by, for example, executing an inference service providing program in the integrated device.

[0254] In the third and fourth embodiments, the information providing device is described as generating the input data. However, the input data may be generated by, for example, an inference device.

[0255] In the third and fourth embodiments, the information providing device has been described as functioning as an acquisition unit, a provision unit, and a billing unit. However, some of the functions realized by the information providing device may be realized on a user's terminal (or on the cloud).

[0256] Furthermore, in the third and fourth embodiments, the details of the method by which the information providing device provides the chemical structure information of lipid molecules to the user are not mentioned, but the method by which the information providing device provides the information is arbitrary. For example, the information providing device may be configured to directly transmit the chemical structure information of lipid molecules to the user, or may be configured to store the chemical structure information of lipid molecules in a storage location that the user can access by entering a password or the like.

[0257] In the fifth and sixth embodiments, the generation unit determines, as a predetermined termination condition, whether the transfection efficiency and / or cell viability inferred by the trained model 520 satisfy the target transfection efficiency and / or cell viability. However, the predetermined termination condition is not limited to this, and may, for example, determine whether the chemical structure information of the lipid molecules to be generated has been updated. In this case, the generation unit determines that the predetermined termination condition is satisfied if it determines that the chemical structure information of the lipid molecules to be generated has not been updated.

[0258] Furthermore, in each of the above embodiments, the information items of FIG. 5, FIG. 10, FIG. 15, etc. are given as examples of information items of input data, but the information items of input data are not limited to these.

[0259] The present invention is not limited to the configurations described in the above embodiments, but may be combined with other elements, etc. These aspects can be changed without departing from the spirit of the present invention, and can be appropriately determined depending on the application form.

[0260] This application claims priority based on Japanese Patent Application No. 2020-146402, filed on August 31, 2020, the entire contents of which are incorporated herein by reference. [Explanation of symbols]

[0261] 100: Drug delivery system review process 170: Data storage unit related to drug delivery systems 200: Drug delivery system review process 210: Learning device 220: Reasoning device 400: Training dataset 520: Trained model 800: Drug delivery system review process 810: Learning device 820: Reasoning device 900: Training dataset 1000: Input data 1020: Trained model 1200: Drug delivery system review process 1211: Information provision device 1400: Drug delivery system review process 1410: Inference service provision system 1420: Reasoning device 1520: Reinforcement learning model 1530: Reward calculation unit 1700: Reasoning device 1710 :Generation part 1900: Reasoning device 1910 :Generation part 1930: Acquisition Department

Claims

1. an acquisition unit that acquires input data including at least chemical structure information of lipid molecules; a trained model generated by performing a training process on a training model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of a nucleic acid encapsulated in a particle containing the lipid molecules into a cell and / or the cell survival rate; An inference device, wherein the trained model infers transfection efficiency and / or cell survival rate associated with input data newly acquired by the acquisition unit.

2. The inference device described in claim 1, wherein the transfection efficiency and / or cell survival rate used when the learning process is performed is calculated from measurement results obtained by introducing nucleic acid encapsulated in particles containing lipid molecules having the chemical structure information used when the learning process is performed into cells.

3. The inference device of claim 2, wherein the trained model is generated by updating model parameters of the training model so that the output when input data including chemical structure information of at least the lipid molecules is input to the training model approaches the transfection efficiency and / or cell survival rate calculated from the measurement results.

4. 2. The inference device according to claim 1, wherein the acquisition unit performs a predetermined preprocessing on the newly acquired input data, and the trained model infers a transfection efficiency and / or a cell survival rate associated with the preprocessed input data.

5. an acquisition step of acquiring input data including at least chemical structure information of lipid molecules; an execution step of executing a trained model generated by performing a training process on a training model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of a nucleic acid encapsulated in a particle containing the lipid molecules and / or the cell survival rate, An inference method in which the execution step executes the trained model to infer transfection efficiency and / or cell viability associated with newly acquired input data in the acquisition step.

6. an acquisition step of acquiring input data including at least chemical structure information of lipid molecules; an execution step of executing a trained model generated by performing a training process on a training model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of nucleic acids encapsulated in particles containing the lipid molecules into cells and / or the cell survival rate, The execution step is an inference program that executes the trained model to infer transfection efficiency and / or cell viability associated with the newly acquired input data in the acquisition step.

7. A model generation method that generates a trained model by performing a learning process on a learning model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of nucleic acids encapsulated in particles containing the lipid molecules into cells and / or cell survival rate.

8. an acquisition unit that acquires input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; A trained model is generated by performing a training process on a training model that associates input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules, The trained model infers chemical structure information of lipid molecules associated with input data newly acquired by the acquisition unit.

9. The inference device described in claim 8, wherein the input data used when the learning process is performed includes the transfection efficiency of an active substance encapsulated in a particle containing a designed or selected lipid molecule into a cell and / or the cell survival rate, calculated from measurement results obtained by introducing the active substance encapsulated in a particle containing the lipid molecule into a cell.

10. The inference device described in claim 8, wherein the trained model is generated by updating the model parameters of the training model so that the output when input data including the prerequisites is input to the training model approaches the chemical structure information of lipid molecules used when the training process is performed.

11. The inference device according to claim 8 , wherein the acquisition unit performs a predetermined preprocessing on the newly acquired input data, and the trained model infers chemical structure information of lipid molecules associated with the preprocessed input data.

12. an acquisition step of acquiring input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; An inference method comprising: an execution step of executing a learned model generated by performing a learning process on a learning model that associates input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules; The execution step is an inference method in which the chemical structure information of lipid molecules associated with the input data newly acquired in the acquisition step is inferred by executing the trained model.

13. an acquisition step of acquiring input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; An inference program for causing a computer to execute the following steps: an execution step of executing a learned model generated by performing a learning process on a learning model that associates input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules; The execution step is an inference program that executes the trained model to infer chemical structure information of lipid molecules associated with the input data newly acquired in the acquisition step.

14. A model generation method that generates a trained model by performing a learning process on a learning model that associates input data containing prerequisites for designing or selecting lipid molecules that make up particles containing an active substance with chemical structure information of the lipid molecules.

15. an acquisition unit that acquires from a user prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; A trained model generated by performing a training process on a training model that associates input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules; a providing unit that provides the user with chemical structure information of lipid molecules inferred by the trained model when input data including preconditions newly acquired by the user is input by the acquiring unit; and An inference service providing system having the above.

16. The inference service providing system of claim 15 further comprises a billing unit that charges the user when the learned model infers chemical structure information of lipid molecules by inputting input data including prerequisites newly acquired by the user by the acquisition unit.

17. The inference service providing system of claim 16, wherein the billing unit changes the billing details for the user when the billing unit obtains from the user the transfection efficiency and / or cell survival rate of the active substance encapsulated in particles containing lipid molecules, the transfection efficiency and / or cell survival rate being calculated from measurement results obtained by introducing the active substance encapsulated in particles containing lipid molecules having chemical structure information inferred by the trained model into cells.

18. An acquisition step of acquiring from a user prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; An execution step of executing a trained model generated by performing a training process on a training model that associates input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules; a providing step of providing the user with chemical structure information of lipid molecules inferred by the trained model by inputting input data including preconditions newly acquired by the user in the acquisition step; An inference service providing method having the steps of:

19. An acquisition step of acquiring from a user prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; An execution step of executing a trained model generated by performing a training process on a training model that associates input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance with chemical structure information of the lipid molecules; a providing step of providing the user with chemical structure information of lipid molecules inferred by the trained model by inputting input data including preconditions newly acquired by the user in the acquisition step; An inference service providing program for causing a computer to execute the above.

20. an acquisition unit that acquires input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including the preconditions acquired by the acquisition unit; a calculation unit that calculates a reward based on the transfection efficiency of the active substance encapsulated in the particle containing the lipid molecule into the cell and / or the cell survival rate, the calculation being calculated from the measurement results obtained by introducing the active substance encapsulated in the particle containing the lipid molecule having the chemical structure information inferred by the reinforcement learning model into the cell, An inference device in which the reinforcement learning model performs learning processing based on the reward calculated by the calculation unit.

21. The inference device according to claim 20 , wherein the calculation unit calculates the reward so that it is maximized by increasing the transfection efficiency and / or the cell survival rate.

22. 21. The inference device according to claim 20, wherein the acquisition unit performs a predetermined preprocessing on the input data, and the reinforcement learning model infers chemical structure information of lipid molecules by receiving the preprocessed input data.

23. an acquisition step of acquiring input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; an execution step of executing a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including the preconditions acquired in the acquisition step; and a calculation step of calculating a reward based on the transfection efficiency of the active substance encapsulated in the particles comprising lipid molecules into cells and / or the cell survival rate, the calculation being calculated from the measurement results obtained by introducing the active substance encapsulated in the particles comprising lipid molecules having chemical structure information inferred by the reinforcement learning model into cells, An inference method in which the reinforcement learning model performs a learning process based on the reward calculated in the calculation step.

24. an acquisition step of acquiring input data including prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; an execution step of executing a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including the preconditions acquired in the acquisition step; and a calculation step of calculating a reward based on the transfection efficiency of the active substance encapsulated in the particle comprising the lipid molecule into the cell and / or the cell survival rate, the calculation step being calculated from the measurement results obtained by introducing the active substance encapsulated in the particle comprising the lipid molecule having the chemical structure information inferred by the reinforcement learning model into the cell, An inference program in which the reinforcement learning model performs a learning process based on the reward calculated in the calculation step.

25. an acquisition unit that acquires from a user prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including prerequisites acquired by a user via the acquisition unit; A providing unit that provides the user with chemical structure information of lipid molecules inferred by the reinforcement learning model; a calculation unit that calculates a reward based on the transfection efficiency of the active substance encapsulated in the particle containing the lipid molecule into the cell and / or the cell survival rate, the calculation being calculated from the measurement results obtained by introducing the active substance encapsulated in the particle containing the lipid molecule having the chemical structure information inferred by the reinforcement learning model into the cell, An inference service providing system in which the reinforcement learning model performs learning processing based on the reward calculated by the calculation unit.

26. a charging unit that charges the user when the providing unit provides the user with the chemical structure information of the lipid molecule inferred by the reinforcement learning model; 26. The inference service providing system according to claim 25, further comprising:

27. The inference service providing system of claim 26, wherein the billing unit changes the billing details for the user when the billing unit obtains from the user the transfection efficiency and / or cell survival rate of the active substance encapsulated in particles containing lipid molecules, the transfection efficiency and / or cell survival rate being calculated from measurement results obtained by introducing the active substance encapsulated in particles containing lipid molecules having chemical structure information inferred by the reinforcement learning model into cells.

28. An acquisition step of acquiring from a user prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; an execution step of executing a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including the prerequisites acquired by the user in the acquisition step; a providing step of providing the user with chemical structure information of lipid molecules inferred by the reinforcement learning model; and a calculation step of calculating a reward based on the transfection efficiency of the active substance encapsulated in the particles containing lipid molecules into cells and / or the cell survival rate, the calculation step being calculated from a measurement result obtained by introducing into cells an active substance encapsulated in the particles containing lipid molecules having chemical structure information inferred by the reinforcement learning model, An inference service providing method in which the reinforcement learning model performs learning processing based on the reward calculated in the calculation step.

29. An acquisition step of acquiring from a user prerequisites for designing or selecting lipid molecules that constitute particles containing an active substance; an execution step of executing a reinforcement learning model that infers chemical structure information of lipid molecules by inputting input data including the prerequisites acquired by the user in the acquisition step; a providing step of providing the user with chemical structure information of lipid molecules inferred by the reinforcement learning model; and a calculation step of calculating a reward based on the transfection efficiency of an active substance encapsulated in a particle comprising a lipid molecule and / or a cell survival rate, the calculation step being calculated from a measurement result obtained by introducing an active substance encapsulated in a particle comprising a lipid molecule having chemical structure information inferred by the reinforcement learning model into a cell, An inference service providing program in which the reinforcement learning model performs learning processing based on the reward calculated in the calculation step.

30. A trained model generated by performing a training process on a training model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of an active substance encapsulated in a particle containing the lipid molecules and / or the cell survival rate; a generation unit that, when the transfection efficiency and / or cell survival rate associated with input data including chemical structure information of a newly generated lipid molecule is inferred by the trained model, repeats a generation process of generating chemical structure information of a next new lipid molecule based on the inference result until a predetermined termination condition is satisfied; An inference device having:

31. The generation unit An inference device as described in claim 30, which selects one of a plurality of search spaces corresponding to combinations of formable hydrocarbon molecular fragments and chemical skeletons of lipid molecules based on the inference result, and generates chemical structure information of the next new lipid molecule using characteristics of the selected search space.

32. The inference device according to claim 31 , wherein the plurality of search spaces differ from one another in combinations of length, degree of saturation, number of branches of molecular fragments, and types of chemical skeletons of lipid molecules.

33. The inference device according to claim 31 , wherein the generation unit generates chemical structure information of the next new lipid molecule under predetermined constraints.

34. Further, an acquisition unit is provided for acquiring prerequisites for designing or selecting lipid molecules constituting particles containing an active substance, The inference device according to claim 30 , wherein the generation unit generates the chemical structure information of the next new lipid molecule using the acquired prerequisite conditions as the predetermined constraint conditions.

35. an execution step of executing a trained model generated by performing a training process on a training model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of an active substance encapsulated in a particle containing the lipid molecules and / or the cell survival rate; a generation step in which, when the transfection efficiency and / or cell survival rate associated with input data including chemical structure information of a newly generated lipid molecule is inferred by the trained model, a generation process for generating chemical structure information of the next new lipid molecule based on the inference result is repeated until a predetermined termination condition is satisfied; A method of inference having the following structure:

36. an execution step of executing a trained model generated by performing a training process on a training model that associates input data including at least chemical structure information of lipid molecules with the transfection efficiency of an active substance encapsulated in a particle containing the lipid molecules and / or the cell survival rate; a generation step in which, when the transfection efficiency and / or cell survival rate associated with input data including chemical structure information of a newly generated lipid molecule is inferred by the trained model, a generation process for generating chemical structure information of the next new lipid molecule based on the inference result is repeated until a predetermined termination condition is satisfied; An inference program that allows a computer to execute the above.

Images